fix(app): pin locale for formatMinutes so unit (linux) isn't runner-dependent

[alltomatos]

Issue for this PR

Closes #236

Type of change

• Bug fix

What does this PR do?

unit (linux) is currently red on dev itself (confirmed on the latest few dev commits, e.g. https://github.com/alltomatos/opencode/actions/runs/34911995346/job/104226783370), which blocks CI on every open PR regardless of what it touches — including #234.

Root cause: formatMinutes() in packages/app/src/pages/stats/stats-controller.ts calls mins.toLocaleString() with no locale argument, so its output silently depends on the process's default locale. Every other numeric/date formatter on the stats page (numFmt, compactFmt, the session date/time formatters) already threads through language.intl() explicitly — formatMinutes was the one outlier. Its unit test hardcoded the pt-BR-style output ("60.327 min", period as thousands separator) that this fork's default locale produces locally, but GitHub's ubuntu-latest runners default to a locale that renders it as "60,327 min".

Fix:

• formatMinutes(ms, locale?) now accepts and forwards an explicit locale.
• stats.tsx's three call sites now pass language.intl(), matching the rest of the file.
• The unit test now pins "pt-BR" explicitly instead of relying on the process's default locale, so it's deterministic on any runner.

How did you verify your code works?

• bun test --conditions=solid --preload ./happydom.ts src/pages/stats/stats-controller.test.ts → 4 pass, 0 fail (was 1 fail before the fix, reproduced locally by not pinning a locale).
• tsgo -b on packages/app shows no new errors from this change (pre-existing errors are all unrelated ghostty-web module-resolution failures caused by a sandbox proxy limitation in this environment, not this diff).
• oxlint on the 3 changed files: 0 errors, pre-existing warnings only.

Screenshots / recordings

N/A — output string format only (now correctly locale-aware instead of relying on the runtime default).

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

---

🤖 Generated with Claude Code

https://claude.ai/code/session_015KrFrdVEmx7apnofWUViJw

---

Generated by Claude Code

[alltomatos]

Confirmed this fix works: @opencode-ai/app#test passes on this PR's own CI. What surfaced next was a second, unrelated pre-existing failure on dev — @opencode-ai/core#test > test/database-migration.test.ts ("declared schema has no ungenerated migrations"), caused by commit 5372b2ed3f never running the migration generator after adding a schema index. Filed a separate fix: #240. Not blocking this PR's own correctness, just noting why unit (linux) may still show red until #240 also merges.

---

Generated by Claude Code

[alltomatos]

e2e (linux) and e2e (windows) are also failing here, but it's the same pre-existing dev regression already tracked in #238 (legacy-new-session.spec.ts:9 and session-timeline-projection.spec.ts:131, reproduced identically on dev's own head and on unrelated PRs) — not caused by this PR's locale fix.

---

Generated by Claude Code