Cardano Updates

Make the Cardano HFC mk-free: Block.hs (CardanoLedgerState + era pattern
synonyms), CanHardFork (EraTranslation diff/values restructure), QueryHF (NS
projection with an era-typed filter), and protocolInfoCardano genesis, which
now produces the per-era genesis Values via the Nary embedding. The canonical
/serialise machinery is stripped from Cardano.Ledger.

There is deliberately no TxIn/TxOut instance for (HardForkBlock xs): tables
are per-era only. With this, lib:ouroboros-consensus:cardano is green.

TODO @js: CardanoTxOut and its IndexedMemPack codecs are retained only
for the deferred snapshot-converter tool.

Update the LSM V2 backend to the mk-free LedgerTablesHandle contract.
lib:ouroboros-consensus:lsm is green.

(The unused containers dependency was dropped from the lsm sublib in
ouroboros-consensus.cabal as part of the abstract-surface commit.)

Update the diffusion layer (Node, NodeKernel, GSM) to the mk-free
ledger/table surface. lib:ouroboros-consensus:diffusion is green.

Finish the mk-free sweep for lib:ouroboros-consensus: the Node entry
points (ProtocolInfo, Run) and the MiniProtocol consumers (ChainSync client
+ InFutureCheck, LocalStateQuery server).

Genesis Values now flow through ProtocolInfo: pInfoInitLedgerTables carries
them and injectInitialExtLedgerState returns (state, values). With this,
lib:ouroboros-consensus is fully green.

Port the mempool to the mk-free MempoolAcc representation: API,
Impl/Common, Query, Capacity, then Update and Init. The mempool no longer
threads a MapKind; it carries the accumulated UTxO values explicitly.

Make the HardFork combinator mk-free: the mk-free state functor and
era-translation types, mk-free extendToSlot, and the SingleEraUTxOHDBlock
split. forward is made config-free (IO pipelining deferred). Implement the
HardForkBlock BlockSupportsUTxOHD bodies plus tick/apply, and the NS type
surface in Ledger.hs, deleting the old canonical machinery.

Flip sweep across the combinator: Query, CommonProtocolParams, PeerSelection,
InitStorage, Forging, Mempool, Embed/{Unary,Nary,Binary}, Degenerate and
Serialisation. Embed/Nary returns the per-era genesis values.

Make the LedgerTablesHandle contract mk-free and fold the InMemory
backend's value ops into BlockSupportsUTxOHD. Drop mk from Storage's
upstream dependencies and make the LedgerDB V2 (StateRef, LedgerSeq,
Forker, snapshots, InMemory backend) mk-free throughout.

Thread LedgerConfig through duplicateWithDiffs/forkerPush and route range
reads through a dedicated EraRangeReader, surfacing an EraRangeReaderProvider
from the V2 LedgerDB. ChainDB consumers are made mk-free, dropping the
vestigial NoThunks/Bridge and CanUpgradeLedgerTables constraints.

Introduce the mk-free abstract table surface (BlockSupportsUTxOHD): the
ledger no longer carries a MapKind type parameter. LedgerState becomes a
plain Type, ApplyBlock is pure, and the dual ledger is made mk-free.

Consolidate the table vocabulary into Ledger.Basics, dropping StateKind and
the now-redundant Ledger/Tables.hs subtree (Tables/{Basics,Combinators,Kinds,
MapKind,Utils}). Haddocks describe the current mk-free API; dead readAll and
stale mk-era references are removed.

This is one atomic semantic change: the lib does not build until the whole
mk-free sweep lands.

autodocodec-schema emits draft-07-compatible schemas, and validators such as ajv accept draft-07 by default. Declaring draft 2020-12 was inaccurate and made ajv fail with "no schema with key or ref .../draft/2020-12/schema". Use the draft-07 $schema URI instead.

Add a test-suite that decodes every example configuration through the autodocodec-derived parsers and runs the full parseConfigurationFiles pipeline on the single-file and split examples. Since the parser shares its codec with the schema, a passing decode is the most reliable validation.

Declare the JSON Schema draft (2020-12) on each emitted schema and turn autodocodec's $comment annotations into description fields, which editors and language servers surface on hover and completion.

Handle --help/-h (printed to stdout), describe each mode, list the available component names, and show how to validate a config file against the emitted schema.

Add a small tool that pretty-prints the configuration JSON Schema: with no argument it dumps the whole-config schema, with a component name it dumps that component's schema, and --list shows the component names. Uses aeson-pretty for readable, key-sorted output.

After the autodocodec port, Protocol builds the per-era genesis NP directly from Data.SOP.Strict (strict-sop-core), so sop-core is no longer used.

Add wholeConfigSchema, which merges the component schemas into a single schema for the configuration's single-file form (every component reads its keys from the same top-level object). The top-level required set aggregates the mandatory keys (the per-era genesis files and the Byron block version).

Add Cardano.Configuration.Schema, exposing a JSON Schema (as an aeson Value) for each configuration component, generated from the autodocodec codecs via jsonSchemaViaCodec. Because they share the codec with the parsers, the schemas cannot drift: they record required vs optional keys, descriptions, and the parse-time defaults. Tracing is excluded, as it is resolved by hermod whose schema lives in trace-dispatcher.

Express each component as a HasCodec instance and derive FromJSON/ToJSON via Autodocodec, replacing the hand-written parsers. Consensus reads the Genesis flags from LowLevelGenesisOptions in GenesisMode; Network covers all peer/diffusion/tx-submission/RPC fields; Protocol builds the per-era genesis NP and Byron/checkpoints from object-codec fragments, reusing the aeson instances for hashes and RequiresNetworkMagic. Drop the now-unused .:= / RelativeFile helpers from Basics and add a shared DiffTime codec.

Port the storage/LedgerDB configuration to autodocodec codecs, and adopt the snapshot model the node is converging on (f-f/prepare-11.1 combined with geo2a/predictable-snapshots): LMDB is gone (backend is V2InMemory or V2LSM), and the snapshot policy lives under a Snapshots key that is either a named preset ("Mithril") or an object of options (SnapshotInterval in slots, SlotOffset, RateLimit, MinDelay/MaxDelay with MinDelay <= MaxDelay, NumOfDiskSnapshots). Only the modern keys are accepted; the deprecated top-level and in-LedgerDB snapshot keys are not.

Introduce autodocodec as the single source of truth for parsing and (eventually) JSON Schema generation. Port the leaf types first: a shared DiffTime codec and File/NodeDatabasePaths/Mempool codecs, deriving FromJSON/ToJSON via Autodocodec. The mempool capacity override keeps its byte-count-or-NoOverride form via optionalFieldWithDefaultWith.

Match the node: the Genesis configuration flags now live under the dedicated LowLevelGenesisOptions key (read only in GenesisMode, where they apply), and EnableCSJ/EnableLoEAndGDD/EnableLoP default to True.

The upcoming node version (f-f/prepare-11.1) removes the LMDB backend entirely and replaces the simple NumOfDiskSnapshots/SnapshotInterval pair with a richer snapshot policy. Mirror that here: the backend selector is now V2InMemory or V2LSM, and the LedgerDB section gains SlotOffset and RateLimit alongside a (now slot-based, non-zero) SnapshotInterval and NumOfDiskSnapshots, with the top-level SnapshotInterval/NumOfDiskSnapshots kept as deprecated fallbacks. Dropping LMDB also lets us remove the live-table path-resolution machinery: the LSM path is stored as parsed and defaulted by the consumer, as the node now does.

Correct the JSON keys to match what the node actually parses: the peer-selection targets are TargetNumberOf*/SyncTargetNumberOf* (not DeadlineTargetOf*/SyncTargetOf*), the LMDB map size key is MapSize, and the mempool capacity override is MempoolCapacityBytesOverride (accepting a byte count or NoOverride). Also accept the deprecated ForkPolicy alias for ResponderCoreAffinityPolicy.

Add example configurations, including a full single-file config and an example of splitting the configuration across per-component files.