runtime: unify mallocStub fast/slow blocks

Inside mallocStub, combine all of the fast path (!isSlowPath_) logic
into a single block, and all of the slow path logic into a single block.

This is primarily a refactor with no semantic change. The intention is
to make it easier to read which part are fast/slow, and to make it
easier for a follow up to change to acquirem early in the fast path.

The only semantic change is to move doubleCheckMalloc earlier. If the
fast path decides to fall back to the slow path, that case will execute
twice, but it is a trivial check and it's even compiled in by default.

This arrangement does require duplicating lockRankMayQueueFinalizer in
both blocks. We can't move it earlier because the fast path falling back
to the slow path would call this twice, which would trigger a lock rank
violation. We could potentially move it below (just before
inlineMalloc). I don't believe it needs to come before
deductAssistCredit, but I'd rather do that in a separate change.

For #79667.
For #79699.

Change-Id: I1101cb112e0c48280c0317d986607ba66a6a6964
Reviewed-on: https://go-review.googlesource.com/c/go/+/794120
Reviewed-by: Michael Matloob <matloob@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
Auto-Submit: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: golang-scoped@luci-project-accounts.iam.gserviceaccount.com <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>