[roll] Roll fuchsia [async-utils] parking_lot -> fuchsia-sync

parking_lot has a few disadvantages for Fuchsia:

* adaptive spin w/ thread yields is great for microbenchmarks
  but has poor system-wide interactions with interrupts and
  the scheduler
* the parking lot abstraction doesn't convey priority inheritance
* implicit heap allocations make it unsuitable for use cases
  like instrumentation code

By contrast, fuchsia-sync is built directly on top of our libsync
and Zircon futexes. It tends to perform slightly worse on
microbenchmarks because it lacks an adaptive spin but feedback
from the kernel team is that we should see better whole-system
performance with this implementation. It's also possible that
changing synchronization strategies will surface new test flakes
in code with latent timing bugs.

Original-Bug: b/42169853
Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/966586
Original-Revision: 23fc11c847901999b041a285bb4c42ee1bdefbbe
GitOrigin-RevId: 32bf2dddcc2ac790f8fc620ee2c8cf43355b78d4
Change-Id: I30e5d20b686e5c78d318eb6c58d2f6c0b3fe2199