refactor(testing): replace dict[str, Any] result bags with typed models

[tcoratger]

Motivation

Two spots returned untyped dict[str, Any] bags with magic string keys, while every other fixture in the package uses a typed model:

• API endpoint handlers returned dict[str, Any] keyed by "expected_status_code", "expected_content_type", "expected_body", read back positionally in the generator. A handler that forgot a key would only fail at the read site.
• The justifiability fixture stored its result as a hand-written camelCase dict ({"delta": ..., "isJustifiable": ...}).

What this does

• Adds EndpointResponseContract(status_code, content_type, body) — the endpoint handler signature is now typed, and a handler that omits a field is caught at construction. This is internal plumbing (the contract feeds the same three ApiEndpointFixture fields), so emitted fixtures are unchanged.
• Adds JustifiabilityOutput(delta, is_justifiable) — the camelCase isJustifiable key now comes from the model alias, and the values are validated. The body/response payloads keep their own API-defined snake_case keys untouched.

Byte-equivalence

Both suites were filled before and after and diffed:

114 passed
diff -rq <golden> <new>  ->  IDENTICAL: no vector churn

The endpoint contract feeds the same fixture fields, and JustifiabilityOutput serializes is_justifiable through the camelCase alias exactly as the hand-written dict did.

Testing

• just check passes (lint, format, ty, codespell, mdformat).
• Fill of tests/consensus/lstar/api + tests/consensus/lstar/state_transition (114 fixtures) is byte-identical to the pre-change output.

🤖 Generated with Claude Code