LATX, AVX: write packed conversions directly to the destination

[luzeng87]

Summary

• write VCVTPS2DQ and VCVTTPS2DQ results directly into the destination vector register
• preserve source values when source and destination alias by scheduling the comparison before conversion
• remove one temporary vector and the final copy

Validation

• clean 3A6000 build: passed
• aliasing and distinct-register cases: passed
• all MXCSR rounding modes, NaN, and overflow cases: passed
• JIT, cold AOT, and hot AOT: passed with a non-empty AOT file
• git diff --check: passed