--refresh-crds is ignored in daemon mode

[ndenev]

Summary

--refresh-crds is handled in batch and interactive modes, but ignored in daemon mode.

Affected code

• Batch refresh handling: src/main.rs (run_batch)
• Interactive refresh handling: src/main.rs (run_interactive)
• Daemon startup path without refresh: src/main.rs + src/daemon/pgwire_server.rs

Problem

Starting daemon with --refresh-crds does not clear CRD schema cache before session initialization.

This is inconsistent with other modes and can leave daemon sessions running on stale CRD schemas.

Why this matters

• Daemon is long-lived, so stale schema effects can persist for a long time.
• Same flag has different semantics depending on mode.
• Users may expect k8sql --refresh-crds daemon to force fresh schema discovery but it currently doesn't.

Proposed fix

1. Centralize refresh logic in one helper (e.g. maybe_refresh_crd_cache(bool)).
2. Invoke it before startup in all modes:
   • batch
   • interactive
   • daemon
3. Keep failures non-fatal (current behavior), but emit warning consistently.

Tests to add

1. Integration/behavioral test that daemon path triggers refresh when flag is set.
2. Regression test ensuring mode parity for --refresh-crds behavior.