switch_context should retry only transient errors, not all anyhow errors

[ndenev]

Summary

switch_context retries on all errors, including permanent failures, because its retry predicate is unconditional.

Affected code

• src/kubernetes/client.rs (switch_context uses retry_with_backoff with |_: &anyhow::Error| true)

Problem

Current behavior retries permanent errors (config/auth/validation) the same as transient network errors. This causes avoidable delay, log noise, and poorer UX.

Why this matters

• Slower fail-fast behavior in REPL/CLI context switching.
• Misleading retry logs for non-retryable failures.
• Unnecessary delay in scripts/automation.

Proposed fix

1. Add retry classification for switch_context path:
   • retry only transient failures (timeouts, transport errors, 429/5xx, temporary unavailability)
   • do not retry permanent failures (invalid context, auth/RBAC, deterministic config errors)
2. Keep exponential backoff for retryable classes.
3. Improve logging to indicate retryable vs non-retryable decision.

Tests to add

1. Permanent error case: verify no retries (or minimal retries as explicitly intended).
2. Transient error case: verify retries occur and respect max attempts.
3. Ensure final aggregated error reporting remains clear across multiple contexts.