metricshandler: stop busy-looping and honor ctx in collector

[tamalsaha]

Problem

StartMetricsCollector ran an unbounded for { ... } loop with two issues (pkg/metricshandler/handler.go):

err := collector.collectMetrics()
if err != nil {
    klog.Errorf(...)
    continue          // skips the time.Sleep below
}
...
time.Sleep(MetricsRefreshPeriod)

1. Busy-loop on error: the continue jumps back to the top without sleeping, so any persistent collection error turns this into a hot loop hammering the API server.
2. Ignores ctx: the loop never checks the passed ctx, so the goroutine keeps running after the manager shuts down.

Fix

• Replace the loop with wait.UntilWithContext, which always waits a full MetricsRefreshPeriod between runs (success or failure) and returns when ctx is cancelled.
• Defensively bound MetricsStore.WriteAll so a partially populated store (fewer families than headers) can't panic the /metrics handler with an index-out-of-range. In normal operation headers and families are built behind the same install flags and stay in sync, so this is hardening rather than a live bug.

Testing

• go build ./... passes.
• go vet and gofmt clean on the changed files.