[roll] Roll fuchsia [starnix] Refactor lockdep guards with tokens

This change refactors `LockDepGuard`, `LockDepReadGuard`, and
`LockDepWriteGuard` to own a `LockLevelToken` instead of manually
managing the lock state. This reduces code duplication and simplifies
the `lock()`, `read()`, and `write()` methods.

It also adds `LockLevelToken` and `assert_lock_level` to allow
asserting lock levels without creating a mutex, and adds tests for these
features.

`LockLevelToken` and `SubclassToken` are now defined in the `tracking`
module, allowing them to be zero-sized types with no-op methods when the
`detect_lock_dep_cycles` feature is disabled, reducing overhead.

Additionally, methods used only internally by these tokens are now
private to the `tracking` module and removed from the dummy
implementation.

Test: fx test starnix_sync-tests
Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/1575156
Original-Revision: d0faa7dbdf36f1374b6b3a52ede69d6faea0e21f
GitOrigin-RevId: c733f0f6903bbf9cb5c8a6f63517fc3e71d5cbc2
Change-Id: I1961128460b43d544fd29d85a66cd98731225e44