Make parallelism optional for singlepass backend

[matklad]

Description

closes #2261

Review

• Add a short description of the change to the CHANGELOG.md file

[matklad]

Something I haven't thought about before writing the code: should this be a compile time or runtime feature? I just brute-codded compile-time, I guess we can add runtime feature if there's a need for that as well, and, even with a runtime feature, ability to not compile rayon is a plus.

In other words, the current solution is what works today, and it's forward compatible with the potentially more general future soln.

[syrusakbary]

This looks great! It should be good to merge, before then can we make this strategy also available in the other two compilers so they all have the same feature?

Thanks!

[matklad]

before then can we make this strategy also available in the other two compilers so they all have the same feature?

I'd rather keep this PR scoped to singlepass backend, as that's my immediate need (and I feel that singlethreadedness makes the most sense for the singlepass) :-)

[nlewycky]

(and I feel that singlethreadedness makes the most sense for the singlepass) :-)

FWIW, I work on all three compilers and I keep a local patch around to make them single-threaded because otherwise I can't debug a bug in any of them. My patch is much worse than yours though, I was adding a Mutex inside the parallel part. Just throwing in my 2¢ that it makes sense for the other compilers too.

[syrusakbary]

bors r+

[bors]

Timed out.

[syrusakbary]

Merging manually, tests are passing