Add support for phase param to custom models

[chrarnoldus]

This is a hacky way purely to benchmark GPT-5.3-Codex. It will be properly implemented in the client when OpenRouter, Vercel AI SDK etc support it.

I am assuming the model will never emit the same text twice in the same task with different phases, but that sounds reasonable to me.

[kilo-code-bot]

Code Review Summary

Status: 7 Issues Found | Recommendation: Address before merge

Fix these issues in Kilo Cloud

Overview

Severity	Count
CRITICAL	0
WARNING	6
SUGGESTION	1

Issue Details (click to expand)

WARNING

File	Line	Issue
src/db/schema.ts	N/A	temp_phase table has no cleanup/retention policy — rows will accumulate indefinitely
src/lib/custom-llm/customLlmRequest.ts	264	undefined taskId stringifies as "undefined" in hash input, risking subtle collisions
src/lib/custom-llm/customLlmRequest.ts	300	Unhandled DB error in streaming hot path — a transient DB failure will crash the user's stream
src/lib/custom-llm/customLlmRequest.ts	708	JSON.parse can throw if init.body is not valid JSON
src/lib/custom-llm/customLlmRequest.ts	792	includeRawChunks was previously gated behind inStreamDebugMode, now always true — increases memory/processing for all requests
src/lib/custom-llm/customLlmRequest.ts	N/A	N+1 database queries — each assistant message in responseCreateParamsPatchFetch computes a key individually, though the batch query mitigates this

SUGGESTION

File	Line	Issue
src/lib/custom-llm/customLlmRequest.ts	N/A	The part parameter name in .filter(part => ...) inside extractMessageTextParts shadows the outer part in convertStreamPartToChunk (not a bug since they're in different scopes, but could be confusing)

Files Reviewed (6 files)

• package.json - 0 issues (openai version bump)
• pnpm-lock.yaml - skipped (generated)
• src/db/migrations/0030_wet_wither.sql - 0 issues
• src/db/schema.ts - 1 issue
• src/app/api/openrouter/[...path]/route.ts - 0 issues
• src/lib/custom-llm/customLlmRequest.ts - 6 issues