Fix time issues

[huitema]

Issue #1601 points out that using gettimeofday() for managing timers can cause errors if the local admin resets that time. It is preferable on Linux to use clock_gettime(CLOCK_MONOTONIC) for that purpose. This requires separating the TLS related usage of time, which has to track "wall time", from the timer-related usage prevalent in QUIC protocol handling.

The solution here is:

• make sure that ticket and token related calls use picoquic_get_tls_time(quic), and rewrite the token and ticket API so as to not take a time parameter;
• incorporate the patch proposed by @fbussery and use CLOCK_MONOTONIC for other protocol actions.

[huitema]

Last point: clock_gettime may in some cases return an error, but merely reading the clock should only cause an issue if the calling arguments are malformed. I would rather ignore the return code than create an error branch that is never exercised.

[fbussery]

p_first_ticket_empty still defined in line 111. Not needed any more

[fbussery]

I have a doubt. In my mind, simulated_time was of same unity as picoquic_get_quic_time which did not change

[huitema]

I also have a doubt. The "gettimeofday" approach has the nice property of inheriting clock synchronization, NTP, etc. This is not strictly required for the transport, but it does make debugging easier, e.g., when comparing time stamps at the client and the server. I wonder whether the use of monotonic clock should be controlled by a compile option.