Refactor transaction label timeout handling and dont leak labels

Problem: It was found that several AVRCP Controller functions would
allocate transaction labels without setting transaction timeouts. This
would cause labels to leak if something went wrong, either in lower
levels of code, or on the remote device. Once our device leaked enough
labels, we could no longer allocate labels for operations, which would
render the connection effectively useless

Solution: We previously added debugging to check which transaction types
were leaking labels. We've used those results with careful code review
to refactor the transaction label code so its much harder to leak
labels. First, context was added to get_transaction, which enables us to
better track which transactions have which labels. Second, the various
build_and_send_* commands now *all* start transaction timers for their
outgoing commands. This means Vendor, Passthrough, and Browsing all have
command timeouts, instead of just Vendor commands. Third, the
transaction timer/timeout code path has been unified into one timer.
Callbacks for browse timeout events have been added as well, so that
upper layers know when something has gone wrong.

Bug: 272585347
Test: Connect a recent device and browse, look at transaction labels in
logs and dumpsys

Change-Id: Ieb73aee5532c06f79265e204e739d8e15c153ae2