fix demangling of inlined frames to not depend on linfo lookup

[vtjnash]

No description provided.

[StefanKarpinski]

Hard to know what this changes or if it's correct since it doesn't have tests. Is it possible to test? If not that's ok, but I think in that case only @vtjnash is in a position to decide if it's ready or not.

[tkelman]

The fixing of the semicolon, and possibly the fromC flag, should address the issue that was causing #19890 to fail its tests on win32, though that PR causes large performance regressions that we will need to address before merging it.

The deletion here of jlsysw_ etc seems mostly unrelated to the rest of what it's doing though?

[vtjnash]

This is hard to test, since it's checking the case where debug info is missing or bad, and we generally try not to do that. The deletion of jlsysw there is because I noticed that its almost dead code (this code path is supposed to be uniform with the other ~3 that compute the same heuristics), but won't actually compute the right set of values if (when) it can get triggered.