Hmm, I think fixing this requires some deeper digging / refactoring of how type error analysis works. It seems like we have some error analyzers that show up as code lenses, and others that are supposed to be quick fixes that don't actually function as such:

E.g. if I have

var x intx := 1

We're supposed to suggest a quick fix changing the last statement to x = 1 (the nonewvars analyzer), but AFAICT this simply doesn't work.

Bumping out of the milestone, as this appears to be a larger project than expected.

Change https://go.dev/cl/511995 mentions this issue: gopls/internal/lsp: don't recompute diagnostics during code actions