Minor GC: restrict global roots scanning to one domain

[Sudha247]

Currently, all live domains scan all the global roots during a minor collection. This patch restricts the global root scanning in minor GC to just one domain. The first domain to reach the scanning part carries on the scanning, with a lock around it. Once it is done, a flag is updated, which indicates that other domains need not scan the global roots.
This is of the patches addressing #424.

The sequential and parallel macro benchmark results for this, are below.

Sequential - Normalised time

Parallel - Time

Results for the Global roots parallel micro benchmark globroots_mp.ml with 10000 roots from ocaml-bench/sandmark#200:

Number of domains	parallel_minor_gc	minor-gc-globroots
1	0.254	0.256
2	0.443	0.439
4	0.556	0.54
8	0.815	0.812
12	1.11	1.039
16	1.88	1.88
20	2.099	2.1
24	2.549	2.518

There is no huge difference between the two branches, this could be due to overheads lying elsewhere.

Examining the eventlogs produced by produce by the same microbenchmark on 12 domains (for a representative value), the event minor_gc/global_roots was found 68081 times in parallel_minor_gc and only 5983 times with this patch.

[ctk21]

LGTM