[Areson]

Stacked on #187: review against Areson/collection-fault-counters.

#187 counts collector-timeouts with errors.Is(err, context.DeadlineExceeded), but some built-in collectors wrap query errors with %s or %v. That flattens the error chain, so when those collectors hit their CMR deadline the timeout isn't counted. This PR changes those wraps to %w. The message text stays the same.

Wraps changed in Collect paths:

• metrics/trx/trx.go: oldest-trx query error (%s → %w)
• metrics/stmt.current/current.go: query error and row scan error (%s → %w, two places)
• metrics/wait.io.table/table.go: rows.Columns() error (%v → %w)
• metrics/repl.lag/pfs.go: replica check query error and lag query error (%s with err.Error() → %w, two places)

The trx and stmt.current changes were flagged in review on #187. I found the wait.io.table and repl.lag ones by grepping metrics/ for %s and %v wraps. Every other match is in Prepare or option parsing, where CMR doesn't apply, so I left those alone.

With this change, the "trx and stmt.current don't yet" caveat comes out of docs/content/api/status.md.

Tests:

• Added mock.ErrorDB (test/mock/db.go). It returns a *sql.DB whose connections always fail with a given error, so tests can check how a collector wraps errors without MySQL.
• Added tests for trx, stmt.current and both repl.lag PFS queries. In each, the query fails with an error wrapping context.DeadlineExceeded, and the test asserts errors.Is still matches after Collect wraps it. With the old wraps, all three fail.
• The wait.io.table Columns() wrap has no unit test. That error only comes from rows that a context cancel closes after the query succeeds, and database/sql closes them in a goroutine, so a test would be racy.
• The stmt.current scan-error wrap has no test either, for the same reason.

Not changed: collectPFS in repl.lag runs its lag query with context.Background() rather than ctx, so CMR doesn't bound that query today. Fixing that changes behavior, so it should be its own change.

🤖 Generated with Claude Code