fix: transmit in single frame when compression enabled

[DanielleMaywood]

Closes #435

Problem

When compression was enabled, Conn.Write sent messages across many small frames due to the flate library's internal bufferFlushSize (240 bytes). Each flush triggered a writeFrame call, producing alternating ~236 and 4 byte frames. This broke clients like Unreal Engine that only process one frame per tick.

Solution

Conn.Write now compresses the entire message into a buffer first, then transmits it as a single frame. Messages below flateThreshold bypass compression and are sent uncompressed in a single frame.

For CompressionContextTakeover mode, the flateWriter destination is restored after buffered compression to ensure subsequent Writer() streaming calls work correctly.

---

🤖 PR was written by Claude Code using Opus 4.5 then reviewed by Opus 4.5 and a human 👩‍🚀

[Copilot]

Pull request overview

This PR fixes a critical bug where compressed WebSocket messages were sent across many small frames instead of a single frame, breaking clients like Unreal Engine that process only one frame per tick. The fix buffers the entire compressed message before transmission.

• Modifies Conn.Write to compress messages into a temporary buffer first, then transmit as a single frame
• Adds logic to send small messages (below threshold) uncompressed in a single frame
• Restores flateWriter destination after buffered compression to maintain compatibility with subsequent Writer() streaming calls

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
write.go	Implements single-frame compression by buffering compressed data before transmission; adds flateWriter destination restoration for context takeover mode
compress_test.go	Adds regression tests verifying single-frame transmission for both compression modes and tests Write/Writer API interoperability

The implementation is well-designed and handles the key edge cases correctly:

1. Mutex handling: The mutex is properly locked once at the start and unlocked via defer in all code paths
2. Buffer management: Uses the buffer pool pattern consistently with proper cleanup via defer
3. Context takeover: Correctly restores the flateWriter destination to ensure subsequent Writer() calls work
4. Error handling: All error paths return properly wrapped errors
5. Test coverage: The tests cover both compression modes, threshold boundaries, and the critical Write→Writer sequence

The code follows the existing conventions in the codebase and the logic is sound. No issues were identified that require changes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[mafredri]

Thanks for the PR!

I've left some comments inline. Main concerns: loss of compression dictionary via Reset, error path safety, and encapsulation.

(PS. Small nit: I'd appreciate it if you also went over comments and evaluated relevance/humanized them.)

[mafredri]

Caught a few more issues 😄

[mafredri]

One last issue, and some minor vanity from my side. 😄