chore: refresh the contract to 26.10.1-SNAPSHOT

[github-actions]

Refreshes the committed contract to 26.10.1-SNAPSHOT, regenerates every client, and
retires the previous version's artifacts and references.

One or more clients FAIL against arcadedata/arcadedb:26.10.1-SNAPSHOT. See the run for which stage.

• @arcadedb/driver (TypeScript): passing
• arcadedb-driver (Python): failing
• github.com/ArcadeData/arcadedb-drivers/go/arcadedb (Go): passing

Closes ##76
#76 (comment)
76

Opened automatically by .github/workflows/contract-watch.yml, force-pushed
in place each day the contract moves.

[claude]

Review

This is a mechanical contract-refresh PR (chore/contract-refresh, opened by contract-watch.yml). Scope-wise it's clean:

• contracts/ still holds exactly one OpenAPI JSON and one .proto, both updated in place (same 26.10.1-SNAPSHOT name, so no adopt-contract-version.sh run was needed, and none of the version-stamped manifests/version.go files needed touching — consistent with the root CLAUDE.md's description of what a same-version refresh does).
• Every other changed file sits under a _generated/generated/gen directory, i.e. nothing hand-edits generated output.
• The one non-generated file touched, python/packages/driver/tests/test_stream.py, is a legitimate test fix (adding "status": 500 to two mocked in-band error payloads) rather than a hand-edit of generated code — needed because NdJsonQueryEventError.message/.status became required fields.

Finding worth a follow-up: the new status/exception/exception_args fields on NdJsonQueryEventError are generated but never read

The contract now declares NdJsonQueryEventError with message: str (previously optional) and three new fields: status: int ("the HTTP status the buffered encoding would have answered the same failure with ... 503 for a retryable conflict, 409 for a duplicated key, 403 ... 413 ... 500 ... Key on this rather than on 'message' to decide whether to retry"), exception, and exception_args. This is exactly the pattern batch_load_stream's error event already has, which the README calls out as "exact, not a fallback" (python/packages/driver/README.md, "Streaming: what the summary carries, and what it does not").

But facade/stream.py::_parse_event (shared by stream_query/stream_command and their async twins) still does:

message = event.error.message
detail = "the stream reported an error" if isinstance(message, Unset) else message
raise ArcadeDBError(200, {"error": detail})

— hardcoding 200 and dropping status, exception, and exception_args entirely, even though they're now always present on the wire. The module docstring's justification ("The raised error's status is 200: honest ... no different from what M4 already established") predates this contract change and is now stale: the contract does carry a real, documented status for this event, the same way it already does for batch_load_stream.

The regenerated test_stream.py is the tell: both updated tests now mock "status": 500 in the error payload, but the assertions right below them are unchanged —

assert caught.value.status == 200

— so the test was patched just enough to satisfy the generated model's new required field, without anyone revisiting whether the assertion it's sitting next to should still read 200. That's not a regression this PR introduces (Go's stream.go and the TypeScript stream.ts hardcode the same 200/"the stream reported an error" convention, and neither was touched here either — TS doesn't validate responses at runtime, which is presumably part of why "TypeScript: passing" while another client is listed as failing), but since all three clients now have a required field they throw away, it seems like a good candidate for a real follow-up: let query_stream/command_stream callers branch on the in-band status for retryability the way batch_load_stream callers already can, and update facade/stream.py's docstring plus the README's "query_stream/command_stream" section (which still says the raised error's status is unconditionally 200) alongside it.

Everything else

• The additive schema changes (AI chat charts, SecuritySeedRequest's admittedPeerId/declaredHttpAddress, the new support/* endpoints, cluster_status additions) are loosely-typed (additionalProperties: true) generated surface with no hand-written facade wrapping them, so no README/CLAUDE.md prose needed updating for those — consistent with how this repo treats unwrapped generated endpoints.
• The PR body is upfront that Python's e2e run fails against the live 26.10.1-SNAPSHOT server; it'd help to know which stage before merging. If it's unrelated to the NdJsonQueryEventError change above, no action needed here — but if it is related (e.g. a real in-band error now reaching the new required status/message fields in a shape the generated model doesn't expect), that's worth confirming first.

Nothing here blocks the mechanical refresh itself — the rules that matter for this kind of PR (generated code left untouched, exactly one contract per kind) all check out.