[runtime] Fix 8, 16, and 32 bit ToInt and ToUint conversions for all cases

[Hans-Halverson]

Summary

The ToInt8, ToUint8, ToInt16, ToUint16, ToInt32, and ToUint32 conversions shared a bug where f64_number.abs().floor() as i64 was prematurely truncating values outside the i64 range. Rewrite these conversions to fix this bug and to be simpler, clearer, and more efficient.

Improvements:

• Fast path covers all smis - even negative values can be directly cast
• Non-smi fast path if rounded f64 is in range for the target int
• Less common finite check and modulo are only applied for non-smis out of the target int range
• For signed conversions we can first cast to an unsigned int of the same width then reinterpret as a signed int

Tests

• Added thorough (LLM-generated) tests for each of these conversions