Use memchr to optimize get_text_slice

[clubby789]

Running the benchmarks on my machine, I get these results:

parse_ctx_runtime/browser                                                                            
                        time:   [106.61 µs 106.73 µs 106.85 µs]
                        change: [-17.886% -17.720% -17.552%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  8 (8.00%) high mild
  1 (1.00%) high severe
parse_ctx_runtime/preferences                                                                            
                        time:   [277.45 µs 277.83 µs 278.29 µs]
                        change: [-8.5355% -8.3601% -8.1949%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  5 (5.00%) high mild
  2 (2.00%) high severe

With some stress tests against Fluent files used in rustc, I saw improvements of 20-30%.

[clubby789]

Some notable improvements when used in rustc:

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.3%	[-0.3%, -0.2%]	3
Improvements ✅ (secondary)	-1.5%	[-4.1%, -0.3%]	41
All ❌✅ (primary)	-0.3%	[-0.3%, -0.2%]	3

[waywardmonkeys]

@flodolo You might want to approve the CI runs for this.

[alerque]

This seems reasonable. I think we're mostly waiting on getting a safe-harbor release out, then we can start moving forward with improvements. Thanks for the contribution.