Rewrite Threading Backend

[ekiwi]

This PR contains a complete from-scratch rewrite of the threading support in chiseltest.

• multi-thread: 2x - 4x speedup on a threading heavy testbench
• single-thread: single threaded testbenches are as fast as (within 3% of) the old, separate single-threaded backend which required an experimental NoThreadingAnnotation.
• Read/write and write/write conflict detection is slightly less strict.
• timescope was removed as a feature
• multi-clock support went from essentially non-existent to really non-existent
• we now have a global step method which just steps the single global clock which is all we support

Benchmark

Single Thread GSD

	New Threaded	Old Threaded	Old No Thread
Treadle	8.8s	77.2s (8.8x)	8.7s (99%)
Verilator	1.9s	58.7s (30x)	1.9s (100%)

UART Multi-Thread Microbenchmark

	New Threaded	New + FSM	Old Threaded	Old No Thread + FSM	vs New Threaded	vs New + FSM
1000 bytes w/ bitDelay=1	0.70s	0.48s (69%)	3.07s (439%)	0.47s	67%	97%
100 bytes w/ bitDelay=50	2.35s	2.31s (98%)	5.67s (241%)	2.23s	95%	97%

+ FSM: Is a version where I manually ported the threads to a naive FSM and thus we only need a single chiseltest thread.