Add fixes to overflow analysis in bounds inference

[rootjalex]

Add fixes (and tests) to bounds inference for the following cases:

• UInt(32) addition and multiplication overflow was not properly detected.
• Division of type.min() by -1 for Int(16) and Int(8) was not detected as overflow.
• Ramps of types that cannot represent their lanes parameter would produce incorrect bounds (e.g. a ramp of bools)

[rootjalex]

@abadams Fixed your comments and also added can_overflow and can_overflow_int functions to Type, I imagine it'll be easier to change things if overflow behavior for different types is changed.

[steven-johnson]

I can open an issue to track this, if deemed worth it!

I'd argue that, as a rule of thumb, all TODOs that are checked in to the codebase should generally have a tracking issue (referenced in the comment). Not always, but usually.

[rootjalex]

Sorry for the delay @steven-johnson . I created #5682 for this problem, and updated the TODOs in 6fa0a56 .

[steven-johnson]

LGTM with nits.

[steven-johnson]

@abadams is this ready to land?

[abadams]

lgtm with two comment nits