[roll] Roll fuchsia [sl4f][bt] fix async deadlocks and await_holding_lock violations

Holding locks across await boundaries prevents the async executor
from making progress and causes thread starvation. In at least
one instance (`get_active_adapter_address`), this pattern was
causing a guaranteed self-deadlock when the stream hit a timeout,
resulting in silent test hangs.

Specific changes:
* Proxy calls: Extracted FIDL proxies using `.clone()` prior to
  `match` blocks, allowing read locks to drop before `.await`ing.
* Watcher streams: Implemented the Extract-Process-Restore pattern
  using `Option::take()` to pull mutable streams out of the global
  state, await them lock-free, and return them afterward.
* Squashed multiple redundant lock acquisitions into single,
  scoped in-place mutations to reduce overhead.

This improves CI stability for SL4F tests, resolves a class of
hard-to-reproduce test timeouts, and burns down the remaining
violations required to enable `#![deny(clippy::await_holding_lock)]`
globally across the tree.

Original-Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/1752269
Original-Original-Revision: 8aabc9a7a9b35c18bc6f1732e6f353ca9074a049
GitOrigin-RevId: abe497dc87cb5210a302dfe65b8bca40958e3e67
Change-Id: I816aba765a404addf096c81b843396dcb6951316