fix(feishu): pass quoted message media to Hermes when replying

[wengxiaoxiong]

Problem

When a user replies to an image or file message in Feishu and @mentions Hermes, the quoted message's media was silently dropped — Hermes only saw the reply text, never the original image or file.

Root Cause

_fetch_message_text() only extracted text from the quoted (parent) message and ignored any media attachments.

Fix

• Added _fetch_parent_message_with_media() which fetches the quoted message and also downloads its images/files via the existing _download_feishu_message_resources() helper, returning (text, media_urls, media_types).
• Replaced _fetch_message_text() call in _process_inbound_message() with the new method.
• Merged parent media_urls/media_types into the current message's media lists so they are forwarded to the agent.
• If the current message is plain text but the parent has media, upgrade inbound_type to the parent's media type (falls back to PHOTO on unknown types).

How to Test

1. In Feishu, send an image to a chat where Hermes is present.
2. Reply to that image message and @mention Hermes with a question (e.g. "what's in this image?").
3. Hermes should now receive and describe the image. Previously it would respond as if no image was present.

Testing

• ✅ Verified manually in a live Feishu workspace — replying to an image message and @mentioning Hermes now correctly surfaces the image to the agent.
• ✅ 6 new unit tests added in tests/gateway/test_feishu.py (TestFeishuFetchParentMessageWithMedia): text message, image message, cache hit, API failure, no-client guard, and end-to-end inbound merge.
• ✅ All 6 tests pass locally.

Platform

Tested on Linux. This change is Feishu-platform-specific (no file I/O, no process management, no cross-platform concerns).

[teknium1]

Thanks for addressing a real Feishu reply-context gap. Current main still fetches only parent text at plugins/platforms/feishu/adapter.py:3259, while the existing resource helper at :3771-3800 can hydrate the parent attachments.

Problems

• gateway/platforms/feishu.py:4018-4020 returns empty media on a text-cache hit. After one reply to a quoted image/file, every subsequent reply to that same parent loses the attachment again; tests/gateway/test_feishu.py:5007-5020 currently expects that loss.
• gateway/platforms/feishu.py:3082-3084 passes MIME strings such as image/jpeg to MessageType; these are not enum values, so all parent media falls back to PHOTO. On current main, PHOTO enters Feishu media batching (plugins/platforms/feishu/adapter.py:3307-3324), changing the semantics of a text reply and misclassifying audio/document/video parents.
• The implementation path was relocated to plugins/platforms/feishu/adapter.py by 560010547.

Suggested changes

• Port this to the bundled Feishu plugin, preserve the triggering reply as MessageType.TEXT, and merge parent media only.
• Store media with the cached parent context (or bypass the text-only cache), then cover repeated replies and image/audio/video/document parents.

Automated hermes-sweeper review.

[teknium1]

parent_media_types[0] is a MIME value (for example image/jpeg), not a MessageType value (photo, audio, etc.), so this always falls into the PHOTO fallback. Please keep the triggering reply as TEXT and let the existing per-attachment MIME routing handle the merged parent media.

[teknium1]

This cache hit drops media for every later reply to the same quoted image/file. Cache the complete parent context including media paths and MIME types, or bypass the text-only cache for this lookup; add a repeated-reply regression test.