fix(spec-drift): reconcile OpenAPI 3.25.0 nightly contract failures (#475)

[TexasCoding]

Summary

Reconciles the SDK to upstream OpenAPI 3.25.0 so the nightly strict contract suite goes green again. Closes #475.

Two hard failures from the scheduled Spec Drift Detection run:

1. MarginMarket.schedule (perps) — new required field referencing nested MarginMarketSchedule (is_open, next_close_ts, next_open_ts). The schedule object is null for 24/7 markets; when present the open/close timestamps are Unix epoch seconds and themselves nullable by phase.
2. POST /portfolio/intra_exchange_instance_transfer (core) — newly listed on the core OpenAPI, marked currently not available. Already implemented on PerpsClient.transfers.transfer_instance (same path/schema). Recorded in _UNIMPLEMENTED_ENDPOINTS rather than dual-shipping a disabled core surface.

Changes

• Vendored specs re-pinned: openapi.yaml 3.24.0 → 3.25.0, plus asyncapi / perps / perps_scm refreshes
• New MarginMarketSchedule model; required schedule: MarginMarketSchedule | None on MarginMarket
• Exported + registered in PERPS_CONTRACT_MAP
• Core transfer path allowlisted in _UNIMPLEMENTED_ENDPOINTS with reason
• Perps market fixtures/tests updated for the required schedule key

Non-goals

• No core KalshiClient implementation of the intra-exchange transfer (endpoint disabled upstream; available on the perps client). Follow-up when Kalshi enables it.
• Soft additive-optional warnings for SCM omitted_subtrader_count left as-is (non-blocking).

Test plan

• uv run pytest tests/test_contracts.py — 666 passed
• uv run pytest tests/perps/ — 507 passed
• uv run mypy kalshi/ — clean
• uv run ruff check on touched files — clean
• CI green on this PR
• Next scheduled nightly run auto-closes Nightly spec-drift: strict contract tests failing (since 2026-07-16) #475 once green

[claude]

Review: fix(spec-drift) — reconcile OpenAPI 3.25.0 (#475)

Reviewed the diff (kalshi/_contract_map.py, kalshi/perps/__init__.py, kalshi/perps/models/__init__.py, kalshi/perps/models/markets.py, tests/perps/test_markets.py, tests/test_contracts.py, plus the four vendored spec files). Note: I wasn't able to execute uv run pytest/mypy/ruff in this review environment, so the below is a static read against the spec diffs and existing conventions — I did not independently re-verify the test-run numbers in the PR description.

Overview

Small, well-scoped drift fix with two independent changes:

1. Adds MarginMarketSchedule and wires it as a required, nullable field on MarginMarket.
2. Allowlists POST /portfolio/intra_exchange_instance_transfer in _UNIMPLEMENTED_ENDPOINTS on the core client, since it's already implemented (same path/schema) on PerpsClient.transfers.transfer_instance.

Correctness

• MarginMarketSchedule modeling matches the spec exactly: is_open: bool (required, non-nullable), next_close_ts/next_open_ts: int | None (required key, nullable value). Leaving these fields without a = None default is the right call — it makes a missing key hard-fail validation rather than silently defaulting, which is what you want for a spec-required field and keeps future required-drift detectable. Good catch documenting why in the comment rather than just what.
• schedule: MarginMarketSchedule | None on MarginMarket (no default) mirrors the same reasoning for the outer nullable-object case (24/7 markets → null).
• Verified kalshi/perps/resources/transfers.py:transfer_instance (sync + async) already targets POST /portfolio/intra_exchange_instance_transfer — the exclusion reasoning in test_contracts.py checks out, and I confirmed there's no duplicate exclusion entry for that path elsewhere.
• Confirmed IntraExchangeInstanceTransferRequest/Response schemas already existed in the core spec pre-diff (not newly added here), consistent with "same path/schema" framing.
• Export lists (kalshi/perps/__init__.py, kalshi/perps/models/__init__.py) and the new _contract_map.py entry all look correctly placed/alphabetized.

Test coverage

• tests/perps/test_markets.py covers both branches well: populated schedule on the happy path (asserts is_open, next_close_ts, next_open_ts roundtrip) and schedule: None for the 24/7 case.
• Minor gap (not blocking): no test exercises the missing key case for schedule or the nested timestamp fields, i.e. nothing directly asserts that omitting a required key raises a validation error. If that's already covered generically by the contract-drift suite (TestRequestBodyDrift/required-field drift tests) rather than per-model, this is fine as-is — worth a one-line confirmation in the PR description if so, since the in-code comments call out "required-drift" as the explicit intent.

Style/conventions

• Follows the project's DollarDecimal/nested-model conventions correctly; model_config = {"extra": "allow"} on the new model matches sibling models in the same file.
• Docstring on MarginMarketSchedule clearly states the nullability contract (which field is null in which phase) — useful given the somewhat unusual "required key, nullable value" spec shape.

Risk / blast radius

• Low risk: additive model change gated behind a required spec field (so any consumer already had to handle this data), plus a test-only allowlist entry for an endpoint the spec itself marks unavailable. No behavior change to request signing, retry logic, or existing public methods.
• The large specs/openapi.yaml/specs/asyncapi.yaml diffs are mostly re-formatting from the vendored re-pin (multi-line YAML descriptions collapsed to single-line) rather than semantic changes — worth a quick sanity check that the re-pin script isn't introducing unrelated formatting churn on every drift PR, since it makes future diffs harder to review, but that's a process nit rather than a defect in this PR.

Nit

• CHANGELOG.md wasn't touched — per CLAUDE.md that's "updated per release," so likely fine to defer to the actual version bump, just flagging in case this was meant to land with one.

Overall: solid, surgical fix that matches CLAUDE.md's spec-first conventions. No blocking issues found.