feat: Query API for SQL over extracted entities (sync spec v0.7.15)

[kdr]

Summary

Syncs the SDK to spec v0.7.15 (cloudglue/cloudglue-api-spec#107) and bumps the package to 0.7.20.

The headline is the new Query API — read-only SQL (or natural language) over the structured data extracted from collections, against the files/entities/segment_entities virtual tables — plus query-call output items on the Response API and an ephemeral thumbnail_url on data-connector file listings.

Generated (cloudglue/sdk)

• New QueryApi: run_query, list_queries, get_query_schema, get_query, cancel_query_export + all Query models
• Response.output items: oneOf message | function_call | query_call
• DataConnectorFile.thumbnail_url (nullable, ephemeral signed provider preview)

Wrapper (cloudglue/client)

• New cg.query resource: run (sql or natural-language), get_schema(collections) (joins IDs into the comma-separated param), list, get, cancel, and wait_for_ready for background exports — registered in client/main.py and resources/__init__

Test plan

• Package imports clean; cg.query wired with all six methods
• Live battery against staging — 12/12 (evidence in ~/Downloads/cloudglue-v0.7.15-sdk-test-results.md): schema introspection lists the three virtual tables + per-collection fields; SQL SELECT COUNT(*) returns inline rows with scan-stat usage; natural-language query returns the compiled SQL; dry_run validates without rows; non-SELECT and face-analysis-collection requests 400; list omits rows while get replays them; background CSV export cancels cleanly; gong/recall listings parse with the optional thumbnail_url

---

Note

Low Risk
Mostly additive OpenAPI regeneration plus a thin client wrapper; no auth or breaking changes to existing resources.

Overview
Regenerates the Python SDK for OpenAPI 0.7.15 and exposes structured querying on the high-level Cloudglue client via cg.query.

Query API adds read-only SQL (or natural-language questions compiled to SQL) over collection extraction virtual tables (files, entities, segment_entities), with schema introspection, stored runs, list/get, background exports (cancel + poll), and dry-run. The hand-written Query resource mirrors the generated QueryApi with run, get_schema, list, get, cancel, and wait_for_ready, wired through client/main.py.

Smaller spec-driven updates: Response.output can include query_call (and related models) alongside messages and function calls; DataConnectorFile gains optional ephemeral thumbnail_url; sync_file_source_metadata docs note iconik poster keyframe thumbnail backfill when missing.

^{Reviewed by Cursor Bugbot for commit 464c047. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want fixes drafted automatically? Bugbot Autofix can create code changes for findings. A team admin can enable Autofix in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 4dcc3be. Configure here.}

[kdr]

Addressed Bugbot's finding in 464c047: query.wait_for_ready now returns the final QueryResult for failed runs instead of raising with only the query ID — the fetched error payload (code/message) is preserved for inspection, matching this client's other wait helpers (e.g. extract.run), and the docstring documents the new contract.

Follow-up verification: the new Response.output items were also confirmed live against staging — a nimbus-002-preview response with entity_backed_knowledge produced three completed cloudglue_query_call items, and the raw payload parses through Response.from_dict with all three accessible via actual_instance. Evidence added to the test-results doc.