perf(cli): guard large-file reads and avoid sync patch I/O

[kilo-code-bot]

Addresses review remarks about memory blowup and event-loop blocking on large file reads.

Changes

• encoding.ts: Cap jschardet detection to a 64 KB prefix. Valid UTF-8 bytes still short-circuit before detection; only non-UTF-8 inputs hit the sampler, so accuracy is unaffected.
• tool/read.ts: Reject files over 10 MB before calling Encoding.read, since the encoding-aware path buffers the whole file for BOM detection / iconv decode (a regression from the prior createReadStream path for multi-GB files).
• patch/index.ts + tool/apply_patch.ts: Convert deriveNewContentsFromChunks to async and replace Encoding.readSync with Encoding.read so the apply-patch flow no longer does blocking I/O on the Node event loop. All callers already live in async contexts.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (2 files)

• packages/opencode/src/patch/index.ts
• packages/opencode/src/tool/apply_patch.ts

---

_{Reviewed by gpt-5.5-2026-04-23 · 242,365 tokens}