feat(feishu): include replied-to message content in agent context

[Cigarrr]

Summary

When a user replies to a specific message in a Lark/Feishu chat, the agent currently has no visibility into the quoted message content — it only sees the user's reply text. This PR fixes that by fetching the parent message via Im.Message.Get and prepending a formatted quote block to the agent's input.

Closes #470

Environment

• Version: 1.2.2-beta.5
• OS: macOS (Darwin 25.4.0, Apple Silicon)
• Agent: Claude Code
• Platform: Feishu/Lark (WebSocket mode)

Changes

• fetchQuotedMessage(parentID): New method that retrieves the parent message content and formats it as [Quoted message from <sender>]:\n<content>\n\n
• extractPostPlainText(content): New helper for parsing Lark rich-text (post) JSON into plain text, supporting both flat and locale-wrapped formats
• onMessage: Captures ParentId from the incoming event and passes it to dispatchMessage
• dispatchMessage: Fetches quoted content when parentID is non-empty, prepends it to text and post message types

Design decisions

• Graceful degradation: Any API failure silently skips the quote — the user's message is always delivered
• Reuses existing patterns: Same Im.Message.Get call used by parseMergeForward, same resolveUserName for sender resolution
• Minimal blast radius: Only platform/feishu/feishu.go modified; no changes to core.Message struct or other platforms
• Scope: Quote injection applies to text and post messages only (the most common reply scenarios). Image/audio/file replies still work as before, just without the quote prefix.

Test plan

• go test ./... passes (pre-existing failure in core package unrelated to this change)
• New unit tests for extractPostPlainText: flat format, locale-wrapped, no title, empty content, non-text tags
• Manual test: reply to a text message in Lark group → agent sees quoted content
• Manual test: reply to a post (rich text) message → agent sees extracted plain text
• Manual test: reply to an image message → agent sees [image] placeholder
• Manual test: normal (non-reply) message → no change in behavior

[chenhg5]

LGTM! Great feature addition for Feishu reply context.

What I like:

• Clean implementation with graceful degradation — API failures don't block the user's message
• Good test coverage for extractPostPlainText (flat format, locale-wrapped, empty, non-text tags)
• Minimal blast radius — only platform/feishu/feishu.go modified
• Reuses existing patterns (Im.Message.Get, resolveUserName)

Edge cases handled:

• Locale-wrapped post content ({"zh_cn": {...}})
• Non-text message types (image, file → [image] placeholder)
• Missing sender info → "unknown" fallback

Thanks for the contribution!

[chenhg5]

This PR has merge conflicts with main because the core fetchQuotedMessage feature was already merged into main (with additional improvements: context.Context propagation, raw API for card support, interactive card extraction, etc.).

I've cherry-picked the remaining unique improvements from this PR's second commit (link tag, hr tag, and improved code_block formatting in extractPostPlainText) into #508, with proper attribution.

This PR can be closed once #508 is merged. Thanks @Cigarrr for the contribution!