storage: reactor stalls during shutdown with 20k partitions

[jcsp]

Seen in 22.1 rc3, but almost certainly not a new issue. Using ManyPartitionsTest from #4529 on i3en.xlarge instances.

Happens after a log line like

34.222.181.103: INFO  2022-05-04 14:36:20,379 [shard 3] cluster - state_machine.cc:139 - State machine for ntp={kafka/scale_000000/7603} handles raft::offset_monitor::wait_aborted (offset monitor wait aborted)

Backtrace is like

 (inlined by) ~metric_groups_impl at /v/build/v_deps_build/seastar-prefix/src/seastar/src/core/metrics.cc:184
 (inlined by) ~metric_groups_impl at /v/build/v_deps_build/seastar-prefix/src/seastar/src/core/metrics.cc:182
~probe at /var/lib/buildkite-agent/builds/buildkite-amd64-builders-i-085d2d03242ede921-1/redpanda/redpanda/vbuild/release/clang/../../../src/v/storage/probe.h:43
 (inlined by) ~disk_log_impl at /var/lib/buildkite-agent/builds/buildkite-amd64-builders-i-085d2d03242ede921-1/redpanda/redpanda/vbuild/release/clang/../../../src/v/storage/disk_log_impl.cc:80
 (inlined by) ~shared_ptr_count_for at /vectorized/include/seastar/core/shared_ptr.hh:464
 (inlined by) ~shared_ptr_count_for at /vectorized/include/seastar/core/shared_ptr.hh:464

This is probably just a case of iterating through destroying thousands of logs, each of which has many sub-object (e.g. metrics) to destroy, and not doing any yields in there.

There's no associated functional issue: the shutdown completes cleanly.