ueventd: allow the ueventd main loop to run in parallel

The coldboot is currently running in multiple subprocesses in parallel
for faster boot time, while the main loop that polls further uevents
runs in a single process and is not parallelized. However, the main
loop also can be the bottleneck of boot, since services such as vold
waits for ueventd to set up device paths and kernel modules. In
addition, initializing kernel modules that take long (e.g. order of
100ms-) in serial takes perceptible time in total.

This change gives ueventd the ability to run the main loop in multiple
threads as well as the cold boot. This feature is configurable by a new
`parallel_ueventd_main_loop` in ueventd.rc. This is a similar syntax to
`parallel_restorecon`, and disabled by default. The number of worker
threads is also configurable by
`parallel_ueventd_main_loop_max_workers`.

Bug: 400592897
Test: DUT boots with and without `parallel_ueventd_main_loop enabled`
Test: Tested `parallel_ueventd_main_loop_max_workers` works
Test: DUT has same SELinux labels for sysfs
Change-Id: I22edd5b84fa502c2a020e4d125af6d95312559c4