fix(python): isolate per-entry parse failures in list_models() (#1302)

[jrob5756]

Fixes #1302 (option 2 — per-entry isolation hardening).

Summary

CopilotClient.list_models() parses the models.list response with a
list comprehension:

models = [ModelInfo.from_dict(model) for model in models_data]

A single malformed entry raises and the whole call fails — the cache stays
empty, so every retry from a fresh session fails identically. The original
multiplier-required ValueError from #1302 is already fixed in
main, but the bot that triaged the issue endorsed implementing
option 2
as a follow-up so future schema drift on individual models can't take down
list_models() for every consumer.

What changed

python/copilot/client.py — wrap each ModelInfo.from_dict(model) in
try/except inside the RPC path:

• Log a WARNING (with model id when available) for each malformed entry.
• Skip and continue parsing the rest.
• Custom on_list_models handlers are unaffected.

Docstring updated to document the new resilience behavior.

Why Python only

• Node.js casts the response (as { models: ModelInfo[] }) — no eager parse.
• Go uses json.Unmarshal — missing fields become Go zero values.
• .NET / Rust rely on their own deserializers with similar semantics.

Only the Python SDK has a hand-rolled per-entry parser that raises on
missing fields, so per-entry isolation is the Python-specific gap.

Tests

New TestListModelsParserResilience class in python/test_client.py with 4 unit tests:

1. test_skips_malformed_entry_and_returns_valid_ones — mixed payload returns only the valid models, with one warning per skipped entry.
2. test_all_malformed_returns_empty_list_without_raising — all-broken payload returns [] instead of raising.
3. test_empty_models_payload_returns_empty_list — empty payload still returns [].
4. test_non_dict_entry_is_skipped_with_warning — non-dict entries (string, int) are skipped without crashing.

Tests use a small _FakeRpcClient stub so they don't need to spawn the CLI subprocess.

Validation

$ pytest test_client.py
56 passed
$ ruff check copilot/client.py test_client.py
All checks passed!
$ ruff format --check copilot/client.py test_client.py
2 files already formatted

ty check reports the same 37 pre-existing diagnostics on
main and on this branch — no new type errors introduced.

Compatibility

• ✅ Backwards compatible — return type is still list[ModelInfo].
• ✅ No public API changes.
• ✅ Custom on_list_models handlers behave identically.
• ✅ Cache semantics unchanged: a successful (possibly partial) RPC response
  is still cached; the cache is still cleared on disconnect.

[Copilot]

Pull request overview

Hardens the Python SDK’s CopilotClient.list_models() parsing so a single malformed model entry in the models.list JSON-RPC response won’t fail the entire call (and leave the cache empty), addressing the resilience gap described in #1302.

Changes:

• Update CopilotClient.list_models() to parse model entries one-by-one, logging a warning and skipping malformed entries instead of failing the whole call.
• Update the list_models() docstring to document the new “skip malformed entries” behavior.
• Add unit tests validating partial/malformed payload handling (including non-dict entries) without spawning the CLI.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
python/copilot/client.py	Implements per-entry try/except parsing in list_models() and documents the new resilience behavior.
python/test_client.py	Adds targeted unit tests using a fake RPC client to verify malformed entries are skipped with warnings.