fix: downgrade DiffFromCache log level for cache-miss errors

[cp319391]

Summary

This PR downgrades the log level for DiffFromCache errors caused by expected Redis cache misses (specifically cache: key is missing) from ERROR to WARN, while preserving ERROR logs for genuine Redis failures.

The diff behaviour and control flow are unchanged; Argo CD continues to fall back to a non-cache diff when the cache is unavailable.

Motivation

In Redis HA / Sentinel setups, cache misses are expected during events such as:

• Redis restarts
• Sentinel master failovers
• Replica resynchronisation

In these scenarios, Argo CD currently logs DiffFromCache failures at ERROR level even though:

• The cache is best-effort and rebuildable
• Applications recover automatically
• No user action is required

This results in repeated error-level log noise and false alerts in otherwise healthy clusters.

What this PR changes

• Downgrades the log level to WARN when the error message contains cache: key is missing
• Retains ERROR level logging for all other Redis/cache errors (e.g. connection refused, timeouts, auth failures)
• Does not change application behaviour, reconcile logic, or diff results

Why this is safe

• Cache misses already trigger a fallback to live diff
• No functional or behavioural changes are introduced
• Only the log severity is adjusted for a known benign case

Scope

• No CLI changes
• No UI changes
• No API changes
• No documentation changes required

Related issues

Fixes: #5068

[bunnyshell]

❗ Preview Environment deployment failed on Bunnyshell

See: Environment Details | Pipeline Logs

Available commands (reply to this comment):

• 🚀 /bns:deploy to redeploy the environment
• ❌ /bns:delete to remove the environment

[ppapapetrou76]

Instead of using a hard-coded value that is vulnerable to future changes, wouldn't it be better to check that the error is ErrCacheMiss?