Make key-rotator rate-limit cooldown configurable per provider

[saivedant169]

What

NewKeyRotator in internal/provider/keyrotator.go takes a rateLimitCooldown, which is how long a key that just hit a 429 is benched before it gets retried. It is hardcoded to a 60s default and there is no way to set it. There is already a TODO on the function saying exactly this (keyrotator.go:35), and every caller passes 0 today (see internal/provider/openai.go:44), so it always falls back to 60s.

Scope

• Add an optional per-provider rate_limit_cooldown field (a duration) to ProviderConfig in internal/config/config.go:520.
• Thread it through to the NewKeyRotator(...) call in internal/provider/openai.go:44 so a configured value is used instead of 0.
• Keep the 60s default when the field is unset or zero, so existing configs behave exactly as they do now.
• Document the field in configs/aegisflow.example.yaml.
• Delete the TODO comment once it is wired.

Acceptance

• Setting rate_limit_cooldown: 30s on a provider benches a 429-hit key for 30s.
• Leaving it unset keeps the 60s behavior, unchanged from today.
• go test ./... passes, with a test that shows a configured value reaching the rotator.

Keep it a plain per-provider duration. Do not expand it into per-key or per-status-code policy, that is out of scope here.

[be-student]

I’m working on this. I’ll add the duration at the provider-config boundary, reject negative values, pass it through buildKeyRotator, preserve the zero/default behavior, and cover the configured cooldown path directly.