Read files in increasing timestamp order in FileRotator

The current design of the readMatching() function reads all
files that match a given prefix, but does not guarantee
the order in which they are read. This could cause
performance problems because the caller is optimized for
inserting files in ascending timestamp order but the files
are read in a reversed order.

This change sorts the list by the timestamp in the file name,
which is more natural since the files are written based on
timestamp. Benchmarking result shows that
testReadFromRecorder_manyUids improved from 1.5s to 1.0s
compares to the worst case.

Also, this is safe because all callers of the
readMatching() function only pass a list of less than 20 files.
None of these callers rely on the order in which the
files are read.

Test: atest ConnectivityBenchmarkTests FileRotatorTest
Bug: 269409485
Change-Id: I73f10890e8f454ee98e7d92fa754d1affe598df4