Summary

The wal collector fails on Amazon Aurora PostgreSQL where
pg_ls_waldir() is not exposed, aborting the scrape every interval
and emitting no WAL metrics.

Root Cause

The collector calls pg_ls_waldir() to count and size WAL segments.
On Aurora the function returns:

ERROR: pg_ls_waldir() is currently not supported for Aurora

The collector treats any scan error as fatal, propagating it up and
killing the collector for that scrape.

Impact

• Aurora users lose all pg_wal_* metrics on every scrape.
• Exporter logs spam the same error every tick.
• This is the WAL-collector analog of Bug: Replication collector crashes on Aurora PostgreSQL #1273 (same root cause, applied
  to a different built-in function).

Expected Behavior

When the underlying function returns the Aurora-specific
feature_not_supported error, the collector should degrade gracefully:
skip emitting pg_wal_segments / pg_wal_size instead of failing the
whole scrape.

Proposed Solution

Detect the Aurora-specific error (PostgreSQL error class 0A
(feature_not_supported) with "Aurora" in the message) inside
collector/pg_wal.go and short-circuit cleanly (skip emitting the
metrics, return without error). This mirrors the approach in #1274
which fixes the same class of bug for the replication collector.

Sketch:

err := row.Scan(&segments, &size)
if err != nil {
    if isAuroraUnsupportedFunction(err) {
        return nil // Aurora doesn't support pg_ls_waldir; skip silently
    }
    return err
}

isAuroraUnsupportedFunction is a tiny helper that checks
pqErr.Code.Class() == "0A" && strings.Contains(pqErr.Message, "Aurora").

Environment

• postgres_exporter v0.19.x / current main
• Amazon Aurora PostgreSQL versions where pg_ls_waldir() is restricted