Fix/haskell string escapes

[theeldermillenial]

Changes

We replace ast.literal_eval() with a custom parser, implement Plutus V3 trailing byte rejection to conform to Haskell validation, and implemented a bunch of TODOs for minor fixes.

Rationale: It's a Python string literal parser that doesn't support Haskell's \DDD (decimal) and \oOOO (octal) escape sequences. Since UPLC string literals follow Haskell conventions (not Python), a purpose-built decoder is more correct. The custom _decode_haskell_string() handles all standard escapes (\n, \t, \xHH, etc.) plus the Haskell-specific ones. No functionality is lost — ast.literal_eval() features like \N{UNICODE NAME} are Python-only and don't appear in UPLC.

This fixes three Haskell compatibility problems in string parsing and script deserialization.

1. Haskell string escape sequences

The UPLC parser now handles Haskell's \DDD (decimal) and \oOOO (octal) string escape sequences, which Python doesn't natively support. Replaced ast.literal_eval() with a custom
_decode_haskell_string() that handles all Haskell escape formats.

Before: (con string "\83\o143") → "\83\o143" (literals kept)
After: (con string "\83\o143") → "Sc" (correctly expanded)

This fixes the string-04 Plutus conformance test case.

2. Strict mode for PlutusV3 trailing bytes rejection

unflatten() now accepts strict=True which rejects programs with trailing data after the flat encoding. The Conway-era PlutusV3 deserializer requires strict mode — any extra bytes after
the flat-encoded program must cause deserialization failure. PlutusV1/V2 remain lenient (default strict=False).

# V3 strict — rejects trailing bytes
program = unflatten(script_cbor, strict=True)

# V1/V2 lenient — ignores trailing bytes (default)
program = unflatten(script_cbor)

Added has_trailing_data() to UplcDeserializer and a finalize() call in unflatten() to check for remaining bits.

3. Minor fixes

• SECP256k1 length checks: verify_ecdsa_secp256k1 validates pubkey (33 bytes), signature (64 bytes), message (32 bytes). verify_schnorr_secp256k1 validates pubkey (32 bytes), signature
  (64 bytes). Matches Haskell's size validation.
• Zero-cost builtin error: budget_cost_of_op_on_model() now raises RuntimeError for builtins not in the cost model instead of silently returning Budget(0, 0). Prevents free evaluation
  of unknown builtins.
• File extension bug: load_network_config() checked file.suffix == "json" but Path.suffix returns ".json". Fixed.

Test results

• 3,997 acceptance tests pass (1 pre-existing failure: missing libsecp256k1)
• 3 new tests for string escape handling
• All 1,246 string-specific acceptance tests pass

[nielstron]

Thanks for your submission! Please add test cases that cover all changed/added functionality (for example the array and boolean builtins). Please also see the individual comments.

[nielstron]

I think removing this will incorrectly lead to accepting invalid programs when specifying the Plutus version. This would lead to incorrect behavior when testing e.g Plutus V1 transactions

[theeldermillenial]

You're right. I've restored PlutusVersionEnforcer in parse() and the UnsupportedTerm handler. The version enforcement at parse time is important for testing versioned programs. I've also added strict mode support to unflatten() as a separate enforcement point for flat-decoded programs (needed for PlutusV3 Conway-era strictness), but that's additive, not a replacement.

[nielstron]

Is this never invoked? I think there should be an error rather than returning a potentially invalid program?

[theeldermillenial]

Restored the UnsupportedTerm handler is back. It was incorrectly removed as part of a broader refactor. The error path is needed when PlutusVersionEnforcer rejects terms during parsing.

[nielstron]

And please make sure to format the files using the pre-commit specified formatter.