Use inline asm! for x86 DIV on Rust 1.59+

[cuviper]

[cuviper]

bors try

[bors[bot]]

try

Build succeeded:

• Format
• Test (1.59.0)
• Test (1.46.0)
• Test (1.31.0)
• Test (beta)
• No Std
• Test (1.34.0)
• Test (stable)
• Test (nightly)
• Test (1.40.0)
• Test (1.36.0)
• Test (i686)

[cuviper]

While this does improve benchmarks here and in my own programs, one downside is that this asm! is opaque to LLVM, so it doesn't get a chance to optimize division by a constant (scalar) value into "magic" multiplication.

[cuviper]

On further reflection, the measured gains should outweigh theoretical optimization losses.

Maybe someday we could try is_val_statically_known to detect divisors that might be optimizable.