pmc: Avoid race conditions in agent update.

The pmc_agent_update() function updates the subscription to
notifications and also the current UTC offset. It uses a timeout of 0
to avoid blocking. When the pmc client sends the first request, the
response from ptp4l may not come quickly enough to be received in the
same run_pmc() call. It then sends the other request and checks for the
response. If it is the response to the first request, it will be ignored.
The update works correctly only if both responses are quick enough to be
received in the same call, or are both slow enough that they are
received in the next call of the pmc_agent_update() function.

The function needs to be called a random number of times in order to
finish one update. If the mismatch between requests and responses
happened consistently, the agent would never reach the up-to-date state
and phc2sys would not enter the main synchronization loop.

Split the update into two phases, where only one thing is updated at a
time. The function now needs to be called at most 3 times to update both
the subscription and UTC offset, assuming it is not interrupted by
another request outside of the agent's update.

Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>