Cardano Updates

Signed-off-by: Chris Gianelloni <[email protected]>

Replace constant memory costs with linear models derived from empirical measurements:

- ValueData: memory = 38×size + 6 (was constant 1)

- UnValueData: memory = 8×nodes + 0 (was constant 1)

  CPU: 290658×nodes + 1000 (was 43200×arg + 1000)

The linear models better reflect actual memory behavior: ValueData scales with serialized size, UnValueData scales with node count.

Benchmark data regenerated with new memory measurement approach.

Update ValueData and UnValueData benchmarks to use createOneTermBuiltinBenchWithWrapper with appropriate memory measurement wrappers (ValueTotalSize and DataNodeCount).

This ensures benchmarks measure the same memory behavior as production builtins.

Apply ValueTotalSize to ValueData and DataNodeCount to UnValueData, replacing plain Value/Data types. This enables accurate memory accounting: ValueData uses total serialized size, UnValueData uses node count for measuring input Data complexity.

Add KnownTypeAst and builtin marshalling instances for DataNodeCount. This enables using the new memory model in builtin definitions while maintaining type safety through the universe system.

Also includes minor refactoring (void instead of (() <$)) for clarity.

Introduce DataNodeCount newtype that measures Data memory via lazy node traversal rather than serialization size. This provides more accurate memory accounting for UnValueData builtin which operates on the Data structure directly without serializing.

The wrapper separates concerns: node counting logic here, cost coefficients in JSON models.

Add new memory-analysis executable with modules for analyzing memory behavior of Plutus builtins. Includes plotting utilities, regression analysis, and experiment framework for deriving accurate memory models from empirical measurements.

Signed-off-by: Roland Kuhn <[email protected]>

try to make windows ci working - part 1

windows fix - part 2

windows fix - part 3

Signed-off-by: Chris Gianelloni <[email protected]>

Signed-off-by: Roland Kuhn <[email protected]>

Apply regression-derived coefficients with overestimation bias.

ValueData: slope 5→38, intercept 0→6 (worst-case model, R²=1.0)

UnValueData: slope 13→8, intercept 16→0 (through-origin, R²=0.995)

Ensures single application of coefficients (JSON only) preventing double-scaling bug.

Replace coefficient-embedded wrappers with simple measurement wrappers. Coefficients now live in JSON cost models for easier updates without recompilation.

ValueData: ValueTotalSize (just returns totalSize)

UnValueData: DataNodeCount (just counts nodes)

Signed-off-by: Chris Gianelloni <[email protected]>