[FEA] Keep the YARN log cache from growing forever #280

Is your feature request related to a problem? Please describe.

On a long-running cluster, the cache keeps information from old ResourceManager logs and can grow without a limit.

Describe the solution you'd like

Add a way to remove old data or limit how far back the tool looks. It must still calculate costs correctly for recent jobs. If someone asks about an older job, the tool should either reload what it needs or clearly say that the job is outside the configured history window.

Test this with many rotated logs, a ResourceManager restart, and both recent and older jobs.

Describe alternatives you've considered

Operators can clear the disposable cache while callers are stopped, but that loses all saved work and does not prevent it from growing again.

Additional context

Follow-up to #278 and the review discussion: #278 (comment)