Avoid very long media metrics deletion blocks

Could see a large block of items expire, which were all done under a
mutex. This blocks other media requests as well as causing long latency
for the current request. Change expiration so we do at most 50 at a
time, so the current thread can respond to its client. If more are due to expire,
fork a separate thread to continue in similar 50-element bursts,
allowing other clients insertion requests to enter between those bursts.

Bug: 109850816
Test: with shorter timeout and limits
Change-Id: I774a0a1bb7fbb81e12b14c540557d23fb055e3f1