Revise bibtex-actions-read to use a dedicated dynamic completion table

[bdarcus]

As a result of various failed efforts to work around limitations in default CRM, really fixing these problems will likely require writing a specialized dynamic completion table that does what we want. Basically, I think it would mean rewriting CRM and using that instead.

See:
minad/vertico#59 (comment)
#155.

To summarize the issues with CRM:

1. terrible for long candidate strings (which we use; see Alternative multiple candidate select UI? radian-software/selectrum#489)
2. initial-input inconsistency (CRM and initial-input question minad/vertico#59)
3. and in general, it just seems a poor step-child to completing-read (though there's probably only so much we can do about that)

So I believe this would mean writing a function to adapt crm to a) shorten that candidates strings to key in the prompt (what I was experimenting with in #155), and b) adding "initial-input" and such to all selections. I don't know how to do either ATM.

Note: one downside is it would only work with fully compliant implementations, like icomplete-vertical and vertico. It currently would not work with selectrum. But in mid-2021, vertico is looking like a better option anyway.

Maybe whether to use single or multiple should be configurable?

It's worth noting that the completing-read-multiple function is fairly short. The entire crm.el file is only about 200 LOC.

[minad]

Alternative #160

[bdarcus]

If the idea in #160 is incorporated into Consult, per that discussion, this will solve the issues with crm (and maybe others).

[bdarcus]

No longer needed with #163.