Allow DTLSv1_set_initial_timeout_duration to be called mid-handshake

This is useful for WebRTC that "guess" a reasonable timeout
duration based on first ICE rtt...but that can be wildly incorrect
so modifying it can reduce worst case behavior.

The patch
a) modifies the timeout for next flight
b) modifies ongoing timer(s).
c) change representation of the DTLSTimer to keep a start time
and a duration rather than only a expire time.

---

Update test as (from davidben@ thanks!)
The tests needed to AdvanceClock() because all the WriteFoo()
methods buffer up a packet, with the other methods implicitly flushing. SetTimeout() was patterned after SetMTU() which was missing an
implicit flush, but probably should have had one, to avoid reordering
of signals.

Once that's fixed, the original test is sensitive to the timeout / 4
business. While I'm here, trim the tests a bit. They included a bunch of
bits that were part of exercising the core ACK machinery. But since
that's already covered elsewhere, we really only need to drive the shim
into a state with an ACK timer and try a SetTimeout

---

Update-Note: DTLSv1_set_initial_timeout_duration now also affect ongoing flights. You might need modify your code to check for timeout after calling the function.

Bug: webrtc:404763475
Change-Id: Iaea911c38e2e7cfc056728d9f47530729032ffc2
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/86167
Commit-Queue: Lily Chen <chlily@google.com>
Reviewed-by: David Benjamin <davidben@google.com>