time.time() should be increasing

[s-cork]

it's possible to get negative timestamp differences in skulpt
But all timestamps should really be monotonically increasing
(unless the system time was changed between calls)

from time import time
t = time()
for _ in range(1000):
  if time() < t:
    print('back in time!')
  t = time()

This pr fixes that with an implementation that entirely depends on the browser performance api.
(falling back to the Date.now() api if the performance api is not available)
The previous implementation combined the two apis which led to negative timings since the fractional offset was not accurate with respect to Date.now().

I've added comments for browser compatibility links to mdn.
I haven't added additional tests since node doesn't support performance without a require('perf_hooks').

---

I also replace this with Sk.global in a few places which I think is safer.

[meredydd]

Er...I'm not sure about this! On a few minutes' reading, it looks as though performance is a monotonic-clock API for performance measurements, not a date/time API, which means that using it like one invites weird effects (such as not updating on a system clock change, or weird readings over suspend/resume, and that's just the things I knew to Google off the top of my head).

[s-cork]

The current skulpt implementation of time.time() already uses performance...
But it combines performance and Date in a way that causes negative time intervals.

I'm happy to just remove performance from time.time() and go with the Date api... But then it's a different pr.
thoughts @meredydd ?

[meredydd]

I think relying entirely on Date makes sense for a datetime module. Let's do that instead.

[meredydd]

Per @s-cork, let's switch to using Date throughout

[meredydd]

Looking good now!