fix(simplex): classify inbound documents as MessageType.DOCUMENT

[lambertian]

Summary

Documents received over SimpleX (PDF / office / zip / .txt / etc.) currently arrive as MessageType.TEXT and are silently dropped. In _handle_new_chat_item, any non-image/non-audio attachment is tagged application/octet-stream and the msg_type block only promotes audio/* -> VOICE and image/* -> PHOTO, while the gateway's document-attachment block (gateway/run.py) forwards attachments only when message_type == MessageType.DOCUMENT. So received documents reach the agent as a text note, not a file.

This adds inbound document classification, mirroring the existing image/voice handling:

• _doc_mime_for() maps common document extensions (pdf/txt/md/csv/json/docx/xlsx/pptx/zip/…) to real MIME types. For anything else it uses mimetypes.guess_type, but collapses any non-application/*/text/* guess to application/octet-stream so a stray image/* or audio/* guess (e.g. .heic, .flac) cannot re-route the file into the vision/STT pipelines.
• the non-image/non-audio branch records that MIME instead of a hard-coded application/octet-stream.
• a new msg_type branch promotes application/* or text/* -> MessageType.DOCUMENT.

Net effect: any attachment that isn't an allowlisted image/audio type is delivered to the agent as a document attachment instead of being dropped.

Tests (tests/gateway/test_simplex_plugin.py)

• test_inbound_document_classified_as_document — a received .pdf dispatches a MessageEvent typed DOCUMENT with application/pdf in media_types. Fails on main (arrives as TEXT), passes with this change.
• test_inbound_unallowlisted_media_routes_to_document_not_photo_or_voice — .heic / .svg / .flac / .mp4 classify as DOCUMENT, never PHOTO/VOICE.
• test_doc_mime_for_maps_known_and_unknown_extensions — extension → MIME mapping, including the collapse and the unknown-extension fallback.

Notes

• Relates to hermes-simplex-bugs-report #30150. This is the inbound receive path; feat(simplex): groups, native attachments, text batching, auto-accept #27978 (native attachments) adds outbound send methods + inbound image/voice but does not type inbound documents — complementary, not a duplicate.
• feat(simplex): groups, native attachments, text batching, auto-accept #27978 also rewrites this same classify region, so whichever lands second will need a trivial textual rebase here.
• Orthogonal pre-existing issue: on daemons that emit fileStatus as a dict, the handler raises before this code runs; not touched here.

[lambertian]

Closing — superseded upstream. The inbound-document classification this PR adds (and the MIME-precision _doc_mime_for helper) both landed on main in v0.17:

• feat(simplex): groups, native attachments, text batching, auto-accept (salvage #27978) #42584 (merged) rewrote the SimpleX adapter, classifying non-image/non-audio inbound attachments as MessageType.DOCUMENT.
• fix(gateway): classify document attachments as DOCUMENT on Signal, Email, and SimpleX #44695 (merged) added the same document-type classification across Signal, Email, and SimpleX, plus the run.py MIME remap (application/octet-stream → guessed MIME via _TEXT_EXTENSIONS + mimetypes.guess_type).

Verified against current origin/main: the inbound classify branch emits MessageType.DOCUMENT for documents, and the gateway remaps the residual octet-stream MIME. Nothing here remains unaddressed upstream, so this branch is no longer needed. Thanks.