The sync engine capped in-flight requests at a hardcoded
MaxRequestsInFlight = 20, applied to file PUTs, deletes, mkdirs and
rmdirs alike. That ceiling is fine for most workspaces but is too
conservative for fast networks / high-tier workspaces (slow uploads of
larger app bundles) and too aggressive when the workspace is
rate-limiting (we want to dial it down without recompiling).

This change:

- Adds Concurrency to libs/sync.SyncOptions, normalised to
  MaxRequestsInFlight in sync.New when unset so existing callers don't
  need updating.
- Threads the limit through groupRunParallel/applyDiff in
  libs/sync/watchdog.go; the constant is kept as the documented default.
- Surfaces a --concurrency int flag on the two user-facing entry points
  that drive this code path: 'databricks sync' and 'databricks bundle
  sync' (the latter is what backs the apps deploy/sync flow).
- Validates --concurrency >= 1 at flag-parse time per the project rule
  on rejecting incompatible inputs early.
- Adds TestGroupRunParallelHonorsLimit to lock in the parallelism cap.

Slack thread:
https://replit.slack.com/archives/C0A2Z9042FR/p1778082184841149