regression in correctness of handling typeinf avoidance

[vtjnash]

These values in the CodeInstance constructor at the linked comment line appear to be mostly set incorrectly now, which can lead to compile corruption (e.g. #51872), as well as now failing to handle the case when retrieve_code_info may fail and return nothing instead of CodeInfo. The old logic had a helper function jl_get_codeinst_for_src which used to handle this case, so that function could either be un-deleted, or that operation could be added as an alternative way to call the CodeInstance constructor, since it needs identical handling to how result would be treated later.

CodeInstance(interp, src, can_discard_trees=)

Originally posted by @vtjnash in #53219 (comment)

Posting as an issue since the PR is already merged so it doesn't get lost