Simplify conditionals upon comparison, rather than instantiation

[weswigham]

Additional work on top of #31354 to retain our better simplification behavior (this is currently set to merge into that PR).

styled-components on #31354:

Files:            30
Lines:         79011
Nodes:        142289
Identifiers:   39324
Symbols:      157061
Types:         95977
Memory used: 182493K
I/O read:      0.03s
I/O write:     0.00s
Parse time:    0.58s
Bind time:     0.24s
Check time:    5.66s
Emit time:     0.00s
Total time:    6.48s

This PR:

Files:            30
Lines:         79011
Nodes:        142289
Identifiers:   39324
Symbols:      157061
Types:         96130
Memory used: 185759K
I/O read:      0.03s
I/O write:     0.00s
Parse time:    0.55s
Bind time:     0.16s
Check time:    5.41s
Emit time:     0.00s
Total time:    6.12s

completionEntryDetails typing delay (via repro from #31302) I sampled at 211ms on #31354 as-is, with this PR I sampled 147ms.

[ahejlsberg]

I was about to commit a very similar change to my PR. I also removed the root-based simplifications from getConditionalType such that simplification only occurs in getSimplifiedType. Opinion on having both? Don't much care for the duplicated code.

[weswigham]

Opinion on having both? Don't much care for the duplicated code.

The only real difference should be in alias symbols - by deferring all simplifications we guarantee that the conditional keeps it's alias, even if it simplifies (since its identity still exists). It's arguably better that way, so yeah, could do that.