fix(gateway): accept html and generic documents

[riccodecarvalho]

Summary

• accept .html/.htm and .tsv as first-class document types with canonical MIME types
• keep generic document uploads accepted instead of forcing users to zip/rename uncommon files
• make text-document context notes explicit that inline content is optional and the saved file should be read when exact source matters
• cover Telegram observed-group document handling and document-context behavior with tests

Verification

• python -m pytest tests/gateway/test_document_cache.py tests/gateway/test_document_context_note.py tests/gateway/test_telegram_group_gating.py -q -o 'addopts='

[alt-glitch]

This was generated by AI during triage.

Related / competing OPEN cluster for inbound document acceptance: #12702 (adds .html/.htm to the allowlist) and #40584 (accepts + inlines small HTML Telegram uploads). This PR is the broadest of the three (generic documents + .tsv + Telegram observed-group handling). Flagging for a maintainer to pick the canonical approach — the allowlist-only and telegram-only PRs may be subsumed by this one.

[teknium1]

Thanks for the gateway coverage. Current main already accepts the PR's generic-document path, but a narrow useful correction remains.

Problems

• Generic authorized-file acceptance is already on main: 4314d451 changed gateway/platforms/base.py:1701-1714 to cache non-media attachments regardless of extension, and Telegram already accepts them at plugins/platforms/telegram/adapter.py:7994-8005.
• The added .html / .htm / .tsv entries in the static expected-extension test are catalog snapshots; tests/gateway/test_document_cache.py's new cache_media_bytes behavior test is the stronger contract.

Suggested changes

• Re-scope salvage to the still-needed pieces: canonical MIME mappings, normalized doc_mime, and the context-note correction. gateway/run.py:2074-2079 currently says every text/* attachment was included, while Telegram intentionally skips inline content above 100 KiB or after UTF-8 decoding fails (plugins/platforms/telegram/adapter.py:8013-8027).
• Keep the behavioral regression test and avoid extending the static extension snapshot.

Automated hermes-sweeper review.

[teknium1]

The behavior test added below is the regression contract for HTML routing. Adding individual extensions to this static list turns it into a catalog snapshot; please rely on the behavior test rather than expanding this enumeration.

[GottZ]

This was generated by AI during triage.

Summary

Two open PRs address adjacent inbound-document classification problems. #51880 adds canonical HTML/TSV MIME handling and corrects misleading document context, while #55777 detects extensionless UTF-8 text and carries explicit inline-state through the Telegram path.

Related pull requests

• #51880 related — (+84/-20) — keep open, narrow before merge: generic-document acceptance is already present on main, but the diff still provides useful canonical MIME mappings for HTML/TSV, normalized Telegram MIME selection, a truthful optional-inline context note, and behavioral coverage. This follows the keep_open review on #51880; remove the redundant generic-acceptance framing and static extension snapshot.
• #55777 related — (+124/-23) — keep open as a separate follow-up: the diff addresses the distinct extensionless-upload cause by conservatively sniffing UTF-8 text, routing the primary Telegram ingress through cache_media_bytes, and tracking whether text was actually inlined. This follows the keep_open review on #55777; its per-attachment inline-state is more precise than #51880's globally optional wording but requires rebasing around the narrower #51880 change.

Suggested consolidation

Merge #51880 after narrowing it to canonical MIME mappings, normalized doc_mime use, truthful context-note behavior, and behavioral tests. Keep #55777 open and rebase it afterward for the distinct extensionless-text classification and explicit inline-state work; neither PR should be closed as a duplicate.

Cross-PR triage: Reviewed 2 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 27 kB of PR diffs, 4 kB of issue/PR text, 5 kB of discussion (4 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.