perf(chat): improve performance of the chat buffer

[olimorris]

Description

I've been meaning to tighten performance on the chat buffer for sometime. It's felt slow when chat history gets big. Every request at a context over 100,000 tokens seems to be delayed by a split second.

I asked Opus 4.6 to conduct a review on the chat class and identify performance improvements. It's findings were:

Priority	Issue	Impact
1	vim.deepcopy(self.messages) on submit	High — scales with conversation length
2	check_context() O(n^2)	High — compounds with many tools/contexts
3	make_safe() called 3x per submit	Medium — builds tables + iterates schema each time
4	sync_all_buffer_content nested iteration	Medium — per-submit cost
5	Chat.new() synchronous initialization	Medium — affects perceived startup time
6	_G.codecompanion_buffers linear scan	Low — only on close, small N
7	Stable context values recomputed	Low — cheap individually

The first 3 were all great catches, which I missed. Possibly because I'm too close to the chat class itself ("can't see the wood for the trees") and the fact I hadn't appreciated the impact of what vim.deepcopy does when it traverses a complex table structure.

For those who are still unsure of what value AI brings to their codebase, I'd reference the commits in this PR and the overall scope which I gave Opus 4.6. I liken it to having an experienced pair of eyes look at my code for what it is with no pre-conceived ideas and biases.

AI Usage

Opus 4.6 (via Claude Code)

Related Issue(s)

Screenshots

Checklist

• I've read the contributing guidelines and have adhered to them in this PR
• I confirm that this PR has been majority created by me, and not AI (unless stated in the "AI Usage" section above)
• I've run make all to ensure docs are generated, tests pass and StyLua has formatted the code
• (optional) I've added test coverage for this fix/feature
• (optional) I've updated the README and/or relevant docs pages