epicsTime: fix overflow handling

[mdavidsaver]

As I noted on lp:1896295, epicsTimeDiffInSeconds() and class epicsTime which underpins it, do not correctly handle 32-bit overflow on targets where sizeof(long)==8 (eg. Linux).

This PR changes 'class epicsTime' to use epicsUInt32 to match 'struct epicsTimeStamp', and tries to fix the various places in epicsTime.cpp where overflow is detected. This passes existing tests, and a new test of epicsTimeDiffInSeconds() overflow.

I'm not certain that this is the best approach though. I think a rewrite would be better. Specifically to make class epicsTime a wrapper around struct epicsTimeStamp.

[mdavidsaver]

Recently, while doing some benchmarks of a driver with perf, I was surprised to see epicsTimeDiffInSeconds() appear as a hot spot. Another reason I'm inclined to rewrite epicsTime*, and simplify it assuming time_t is integer seconds (as it is on all known targets).

[AppVeyorBot]

❌ Build EPICS Base 7 base-7.0-100 failed (commit 9bdb582fea by @mdavidsaver)

[mdavidsaver]

Superseded by #185