Add string to double parsing benchmarks

Summary:
This benchmarks various functions that convert decimal values in string representation to IEEE-754 double representation.

These benchmarks `std::strtof`, `std::strtof` with a copy (needed for non-c-strings), `std::strtofl` with the C locale (for locale-indepent processing), `std::from_chars`, libdouble-conversion, and libfast_float.

The functions are benchmarked on different inputs. There is input that is hardcoded values in decimal and exponentional notation. There is randomly generated values in the double space range. There is also inputs for single and double digit ints. As well as percentages.

Note that `std::from_chars`, which on platform010/libstdc++ v11 uses `strtof` + `uselocale` [0]. libstdc++ v12 uses fast float[2].

0: gcc-mirror/gcc@932fbc8
1:  gcc-mirror/gcc@490e230#diff-d3c32d9c9c566f7f3888d150c6448428ea194170146a1a166917ba45b1252187

Reviewed By: yfeldblum, Orvid

Differential Revision: D61356955

fbshipit-source-id: 6ec21b602b08505d946551dda49a35402bee7dae

build/fbcode_builder/manifests/fast_float

@@ -0,0 +1,20 @@
+[manifest]
+name = fast_float
+
+[download]
+url = https://github.com/fastfloat/fast_float/archive/refs/tags/v2.0.0.tar.gz
+sha256 = 5d528ec20811577c5f2b2873528c085c500fdcd2b2c0901450509a71de5f1fa4
+
+[build]
+builder = cmake
+subdir = fast_float-2.0.0
+
+[cmake.defines]
+FASTFLOAT_TEST = OFF
+FASTFLOAT_SANITIZE = OFF
+
+[debs]
+libfast-float-dev
+
+[rpms.distro=fedora]
+fast_float-devel

build/fbcode_builder/manifests/folly

@@ -20,6 +20,7 @@ libdwarf
 libevent
 libsodium
 double-conversion
+fast_float
 fmt
 lz4
 snappy