Cardano Updates

Re-run string builtin benchmarks with the unified 4-byte worst-case
generator and fit new cost model parameters. Add cost model variants
D and E that use byte-based costing via TextCostingByteLength.

Update conformance test budgets to reflect the new model parameters.

Replace per-function text generators with a unified 4-byte character
generator that produces worst-case inputs for all string builtins.
This ensures benchmark measurements reflect the most expensive case
(4-byte UTF-8 characters) rather than an ASCII-biased average.

Extract shared Nop overhead subtraction into Common module to reduce
duplication across benchmark suites.

Add EncodeUtf81Byte, AppendString1Byte, and EqualsString1Byte benchmarks
using ASCII text generators. These produce a second data series for
costing analysis, confirming that byte-based sizing gives consistent
costs regardless of character width.

Add interactive cost model visualization pages for AppendString,
EqualsString, EncodeUtf8, and DecodeUtf8 with configurable data
sources (branch selector, custom CSV/JSON URLs). Includes a local
dev server (serve.sh) that serves docs with /data/ mapped to the
local cost-model data directory.

Fix axis labels from "byte-words" to "characters" for string builtins
that take Text inputs (AppendString, EqualsString, EncodeUtf8).

Introduce TextCostingByteLength newtype that measures Text memory by
UTF-8 byte length (in 4-byte half-words) instead of character count.
This correctly reflects that ASCII text is 4x cheaper to process than
worst-case 4-byte characters.

String builtins (AppendString, EqualsString, EncodeUtf8, DecodeUtf8)
now use variant-aware denotations: semvar A/B/C retains the original
character-count costing, while semvar D/E uses byte-based costing.
Golden signature tests are split per-variant accordingly.

Updated sizes reflect smaller generated code from using (<>) directly
for lexicographic ordering in TH-derived compare implementations.

Move existing deriveEq golden files into 9.6/ subdirectories and
wrap tests with testNestedGhc for GHC-version-specific output.

Add Spec.Ord.Golden with golden tests for deriveOrd on Extended and
ProtocolVersion. Wire into the test suite.

Add Ord.Spec with golden tests for TH-generated code and property
tests verifying Ord laws. Wire into the test suite. Use testNestedGhc
for GHC-version-specific golden file organization.

Replace the hand-written Ord instance for Extended in V1.Interval
with a TH-derived one. Add deriveOrd for ProtocolVersion in
V3.Contexts.

Replace the hand-written Ord instance for These with a TH-derived
one, matching the existing deriveEq pattern.

Add Ord.Class and Ord.TH modules that provide a TH-based deriveOrd
analogous to deriveEq. Refactor PlutusTx.Ord to use deriveOrd for
all standard type instances (tuples, Bool, Maybe, Either, etc.)
instead of hand-written implementations. Add Monoid instance for
Ordering (needed by (<>) in generated compare code). Add HLint
suppressions for Ord.Class.