fix(ptodsl): support runtime float integer casts

[Zhendong404]

Summary

• Add runtime scalar float-to-integer casts through arith.fptosi and arith.fptoui.
• Add runtime scalar integer-to-float casts through arith.sitofp and arith.uitofp.
• Confirm and cover scalar.cast in ordinary scalar logic, outside SIMT execution.
• Reject float builtin vector to integer casts with a clear TypeError instead of constructing invalid MLIR.
• Support explicit index -> float conversion through arith.index_castui and arith.uitofp.
• Document truncation-toward-zero and the undefined/poison behavior for out-of-range, NaN, and infinity float-to-integer inputs.
• Add coverage for float literals, index-to-float, f16/bf16 and i8/i16/i64 widths, and fractional ST inputs.

Validation

• ninja -C builds/issue-1179-scalar-cast check-dsl — 33/33 PTODSL tests passed.
• scripts/sim_dsl.sh test/dsl-st/scalar_cast.py — all 9 ST cases passed on simulator, including fractional inputs that verify truncation toward zero.
• python ptodsl/tests/test_scalar_cast.py — passed.

Closes #1179

[Zhendong404]

Should be covered by #1189