Revert #44101, revert #44097, and revert #43852

[DilumAluthge]

I now believe that the recent CI breakage had nothing to do with the public https://badssl.com/ instance. Instead, I believe that the breakage was caused by one (or more) of the following pull requests:

1. Take purity modeling seriously #43852 (f090992)
2. inlining: use ConstResult if available #44101 (fed4544)
3. Fix doctests for CodeUnits indexing. #44105 (0bf6ce3)

This PR reverts the above three PRs.

[Keno]

#44097 is unrelated

[Keno]

#44105 is related though, but only affects doctests so this should be enough to test your hypothesis.

[DilumAluthge]

#44105 is related though, but only affects doctests so this should be enough to test your hypothesis.

Yeah I just noticed that the doctests are failing, but yeah I agree, let's first see what the testers say.

[DilumAluthge]

@Keno Except for the doctests, all CI jobs are passing on d21e8eb (https://buildkite.com/julialang/julia-master/builds/8632)

I'm going to remove the revert of #44097 and add a revert of #44105.

[DilumAluthge]

If all CI jobs pass, I'd be fine with merging this PR, with the plan of re-landing the three reverted PRs once the underlying bug is fixed.