remove build script, bump MSRV to use standard cfg attributes for SIMD

[CathalMullan]

Resolves #208, Resolves #125, Resolves #60

• Removed the build script.
• Bumped MSRV to 1.59 in order to guarantee access to NEON intrinsics.
• Update to 2021 edition (only required some macro syntax tweaks).
• Replaced httparse_simd_target_feature_sse42, httparse_simd_target_feature_avx2, httparse_simd_neon_intrinsics with target feature checks.
• Kept httparse_disable_simd option to disable SIMD entirely, falling back to SWAR. This now has to be configured via RUSTFLAGS, rather than CARGO_CFG_HTTPARSE_DISABLE_SIMD.
• Removed httparse_disable_simd_compiletime option to force runtime detection. Can always force runtime detection by not setting any target_feature anyways.
• Disable SIMD when running under miri.

I copied the existing behavior as close as possible when it comes to the feature flags.
That is, when "std" isn't enabled, we always fallback to SWAR.
Even though it looks like compile-time SIMD could work fine without "std".

Also noticed that there was a bug in the CI benchmark for SSE4.2.
It previously tried to set "sse42", rather than "sse4.2".

This feels like it probably should be considered a breaking change, but I guess that depends on whether the cfg options in the build script were considered part of the public API or not.

There are 2 follow up tasks worth looking into:

• remove the "std" gating on compile-time SIMD support, since only runtime detection needs "std"
• simplify the runtime detection (as mentioned in Can the build script be removed or reduced? #208, assuming benchmarks look good)

[seanmonstar]

Phenomenal work, thank you thank you!

I don't consider any of this a breaking change, the parts detect in the build script were implicit, not public promised.