Fix stability and resource hygiene regressions

[Rohithmatham12]

Summary

• add an idempotent FragmentTable Close method and stop the scheduler fragment-table cleanup goroutine during scheduler shutdown
• make the query-eviction metric test wait for the async eviction counter increment before asserting
• stop comparing live rule evaluation timestamps/durations against never-evaluated backup rules in ruler backup tests
• make the gzip decompression-cap regression test use a valid oversized protobuf payload so removing the cap would make the test pass incorrectly

Issues
Fixes #7596.
Fixes #7604.
Fixes #7577.
Fixes #7581.

Testing

• GOCACHE=/tmp/cortex-go-cache GOMODCACHE=/tmp/cortex-go-mod-cache go test -tags "netgo slicelabels" ./pkg/scheduler/fragment_table ./pkg/util/queryeviction ./pkg/util -run 'Test|TestPrometheusMetrics_IncrementedCorrectly|TestParseProtoReader_GzipDecompressionBomb'
• GOCACHE=/tmp/cortex-go-cache GOMODCACHE=/tmp/cortex-go-mod-cache go test -tags "netgo slicelabels" ./pkg/ruler -run 'TestGetRulesFromBackup|TestGetRules_HA' -count=20
• GOCACHE=/tmp/cortex-go-cache GOMODCACHE=/tmp/cortex-go-mod-cache go test -tags "netgo slicelabels" ./pkg/scheduler ./pkg/scheduler/fragment_table
• GOCACHE=/tmp/cortex-go-cache GOMODCACHE=/tmp/cortex-go-mod-cache go test -tags "netgo slicelabels" ./pkg/util/queryeviction -run TestPrometheusMetrics_IncrementedCorrectly -count=200
• git diff --check