fix(config): keep read paths consistent with OCR_CONFIG_PATH

[dvd233]

Description

Closes #1484.

OCR_CONFIG_PATH already controls ocr llm test, but review, scan, and
preview loaded app settings from the default home config instead. Provider
wizards also saved to their explicit config path and then invoked the
environment-based test command, so a set OCR_CONFIG_PATH could test a
different file from the one just saved.

This change:

• uses resolveConfigPath() for review/scan runtime setup and preview token resolution;
• passes the just-saved path to manual, custom, and built-in provider wizard connection tests;
• keeps ocr config set and provider wizard writes on defaultConfigPath();
• adds regression tests for override reads, preview settings, writer isolation, and wizard test-path selection;
• documents OCR_CONFIG_PATH in the English configuration guide.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Refactoring (no functional changes)
• Documentation update
• CI / Build / Tooling

How Has This Been Tested?

• CC=gcc CGO_ENABLED=1 go test -v -race -count=1 ./...
• CC=gcc CGO_ENABLED=1 go vet ./...
• gofmt on all changed Go files and git diff --check
• bash scripts/verify-license.sh
• go run scripts/verify-english-only.go
• go test -count=1 -coverprofile=... ./... — 91.1% total coverage
• go build ./cmd/opencodereview
• make check — equivalent checks pass; the local BusyBox make invokes the Bash license script with an incompatible /bin/sh
• Manual LLM review — not run because this environment has no configured LLM endpoint

Checklist

• My code follows the project's coding style (gofmt, go vet)
• I have performed a self-review of my code
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally with my changes
• I have updated the documentation accordingly
• I have signed the CLA
• I used AI/LLM assistance and disclose it below; I reviewed the output, did not attribute the commit, and will answer maintainer questions myself.

AI disclosure

Prepared with OpenAI Codex through pi using model gpt-5.6-luna. The generated
code and text were reviewed manually, and the commit contains no AI attribution
trailer.

Related Issues

Closes #1484.

[github-actions]

✅ OpenCodeReview: Review complete: 0 finding(s) across 4 selected item(s).

[lizhengfeng101]

@dvd233 Thanks for the fix, and for the clear write-up.

The wizard part is right and I want to keep it: having applyManualConfig / applyCustomProviderConfig / applyOfficialProviderConfig test the path they just wrote is correct regardless of what happens to the environment variable, and the runLLMTestPath seam is a clean way to get there.

On the direction, though, I'm going with option (b) from #1484 — dropping OCR_CONFIG_PATH — rather than (a). Three things decided it:

It was never a public interface. It is not in the README (any locale), not on the docs site, not in --help. It landed in #204 purely so the VS Code panel could test a draft config before saving. Removing it is not a breaking change and needs no deprecation window.

Its only two callers do not need it. Both extensions/vscode/.../CliService.ts and extensions/idea/.../CliService.kt already set HOME and USERPROFILE to the temp dir before running ocr llm test. defaultConfigPath() resolves through os.UserHomeDir(), so the draft config is already found without the variable. It has been redundant since the day it was added.

Widening it has a cost I don't want to pay. Making loadLLMRuntime honour it means ocr review takes its config — and therefore mcp_servers.<name>.setup (executed via shellCommand, review_cmd.go:568) and api_key_cmd / auth_token_cmd (executed via sh -c, keycmd_unix.go:33) — from a path chosen by an environment variable. The env overrides we have today (OCR_LLM_URL and friends) have no _cmd variant, so they cannot execute anything. ocr review is the command that runs in CI, and I would rather it not gain that property. The comment this patch rewrote was recording exactly that trade-off, even though it framed it as being about writes.

Would you like to revise the PR along those lines? Concretely:

• keep the runLLMTestPath seam and the three wizard call sites;
• drop resolveConfigPath(); runLLMTest() goes back to defaultConfigPath();
• remove the configPath parameter from testConnection in both extensions;
• drop the now-dead env plumbing in config_cmd_test.go and provider_cmd_test.go;
• no docs change needed.

Either way is fine — if you would rather not, say so and I will take it from here; the wizard fix stays yours.

One housekeeping item: the CLA checkbox is still unchecked, and that needs signing before I can merge.