Overhaul Unicode preferences

[lgoettgens]

To be able to disable unicode for tests/doctests etc., see e.g. oscar-system/Oscar.jl#3271

Resolves #1588.

[codecov]

Codecov Report

Attention: 22 lines in your changes are missing coverage. Please review.

Comparison is base (14b8c2b) 87.17% compared to head (6108fee) 87.14%.

Files	Patch %	Lines
src/PrettyPrinting.jl	0.00%	22 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1586      +/-   ##
==========================================
- Coverage   87.17%   87.14%   -0.04%     
==========================================
  Files         115      115              
  Lines       29565    29577      +12     
==========================================
+ Hits        25773    25774       +1     
- Misses       3792     3803      +11     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lgoettgens]

This now contains the sketch by @fingolfin in #1588 as well with one difference: with_unicode does not longer call allow_unicode to be able to have multiple with_unicode stacked onto one another.
Furthermore, I noticed that calling allow_unicode from within a with_unicode block may have unintended results (due to the with_unicode finally clause). This is both in the current and in the updated form. I added this fact to the docstring of allow_unicode.

[thofma]

No clue (neither what the problem nor what the proposed solution is).

Does that address your concerns @benlorenz?

[fingolfin]

Some optional suggestions added

[fingolfin]

Seems we have no tests for with_unicode in AA` :-(. But of course this is out of scope for this PR.

[lgoettgens]

That's true. But after thinking again about it, this seems to be kind of hard anyway:
Calling allow_unicode() from the test would change the user preferences while running tests (not a great experience for devs), and from a within with_unicode block allow_unicode is unsafe.

But overall, we should postpone it

[lgoettgens]

All of this is non-breaking and can thus be merged, right?

[benlorenz]

I would also prefer to have this as a const Ref, this should work with all supported julia versions. I think this avoids some performance penalty in the compiler since it can generate more specific code and it also avoids having to specify the global every time. I added suggestions for the remaining code.

Otherwise I think this should work for the Oscar tests.