Fix two compile-time bugs

[eeckstein]

• Comment out an expensive assert in type lowering

For very large types (e.g. tuples with many elements), this assertion is very expensive to execute, because the isLegalSILType status is not cached.

• LICM: add limits for the number of alias checks to avoid large compile times in corner cases

Avoids quadratic complexity in corner cases. Usually, these limits will not be exceeded.

Resolves #62837

[eeckstein]

@swift-ci test

[eeckstein]

@swift-ci test macos

[eeckstein]

@swift-ci test macos

[atrick]

@eeckstein I don't have any big concern with these changes. But loops didn't become bigger and types didn't become bigger. So it would be interesting to know why asking for a lowered SIL type or querying AA suddenly caused the compiler to "hang".

[eeckstein]

@atrick The main problem was the assert in getTypeLoweringForLoweredType. The old escape analysis had its own implementation for checking if a type is a "pointer" type. So it didn't use this code.
I believe, the problem in LICM always caused compile time problems, but we didn't notice so far.

[atrick]

@eeckstein I wonder why that assert was checked before the cache lookup. It might be beneficial to run after a cache failure.