Fix backup and restore reliability bugs

[itbm]

Summary

Fixes the reliability bugs identified in review (except the intentional dual image names).

Changes

1. S3 checksum failures — default AWS_REQUEST_CHECKSUM_CALCULATION / AWS_RESPONSE_CHECKSUM_VALIDATION to when_required to avoid XAmzContentSHA256Mismatch with AWS CLI v2 / S3-compatible stores
2. Dump cleanup — write dumps under /tmp and remove them after upload (and on failure via trap)
3. Ash compatibility — replace [[ … ]] / == with POSIX case / = so scripts work under Alpine BusyBox ash
4. UTC timestamps — use date -u so the Z suffix is accurate
5. Encryption — encrypt with PBKDF2 (100k iterations); restore tries PBKDF2 then falls back to legacy OpenSSL KDF
6. POSTGRES_EXTRA_DUMP_OPTS — dump-only flags for pg_dump / pg_dumpall without affecting restore clients
7. Cron timeout — remove the hard-coded 1h limit; optional COMMAND_TIMEOUT (go duration, e.g. 2h)
8. Overlapping schedules — skip a run if the previous command is still running
9. SQL identifier safety — quote database names for DROP / CREATE DATABASE

Docs

• README updates for the new env vars and PBKDF2 decrypt command

Test plan

• dash -n on backup.sh, restore.sh, run.sh
• go build for go-cron
• OpenSSL PBKDF2 encrypt/decrypt round-trip
• Legacy OpenSSL decrypt fallback
• Overlap skip smoke test with @every 1s