For the 2 new functions I've followed current naming scheme, replacing
"lsp-abiding" by "bytewise".  We can then get rid of all the suboptimal
nomenclature (lsp-abiding, column, bytewise) in one go at a later stage.

I also didn't attempt to make 'eglot-move-to-bytewise-column' and
'eglot-bytewise-column' "correct" in the cases where 'json-serialize'
throws an error.  I think it's clear that Eli and I have different
definitions "correctness" in this context, but Eli's proposal requires
substantial further work, including recovering from possible errors
raising from 'json-serialize'.

Lastly, I used 'line-beginning-position' naively in the 2 new function
because there are at least 4 other pre-existing naive uses in eglot.el.
This refinement deserves a patch of its own, probably using something on
the lines of

  (defalias eglot--pos-bol (if (fboundp 'pos-bol) 'pos-bol ...))