MSVC optimizations for count_digits. (#1890)

Changed the clz implementations to use xor instead of subtraction so that when
count_digits "undoes" the BSR -> CLZ translation, the optimizer is more
willing to recognize the equivalence.
Changed the data array in bsr2log10 to static since otherwise MSVC generates
code to build the array every time the function is called.

include/fmt/format.h

@@ -211,7 +211,7 @@ inline int clz(uint32_t x) {
   // "r", but the only way that can happen is if "x" is 0,
   // which the callers guarantee to not happen.
   FMT_SUPPRESS_MSC_WARNING(6102)
-  return 31 - static_cast<int>(r);
+  return 31 ^ static_cast<int>(r);
 }
 #  define FMT_BUILTIN_CLZ(n) detail::clz(n)
 
@@ -225,13 +225,13 @@ inline int clzll(uint64_t x) {
   _BitScanReverse64(&r, x);
 #  else
   // Scan the high 32 bits.
-  if (_BitScanReverse(&r, static_cast<uint32_t>(x >> 32))) return 63 - (r + 32);
+  if (_BitScanReverse(&r, static_cast<uint32_t>(x >> 32))) return 63 ^ (r + 32);
   // Scan the low 32 bits.
   _BitScanReverse(&r, static_cast<uint32_t>(x));
 #  endif
   FMT_ASSERT(x != 0, "");
   FMT_SUPPRESS_MSC_WARNING(6102) // Suppress a bogus static analysis warning.
-  return 63 - static_cast<int>(r);
+  return 63 ^ static_cast<int>(r);
 }
 #  define FMT_BUILTIN_CLZLL(n) detail::clzll(n)
 
@@ -901,7 +901,7 @@ template <typename T = void> struct FMT_EXTERN_TEMPLATE_API basic_data {
 // Maps bsr(n) to ceil(log10(pow(2, bsr(n) + 1) - 1)).
 // This is a function instead of an array to workaround a bug in GCC10 (#1810).
 FMT_INLINE uint16_t bsr2log10(int bsr) {
-  constexpr uint16_t data[] = {
+  static constexpr uint16_t data[] = {
       1,  1,  1,  2,  2,  2,  3,  3,  3,  4,  4,  4,  4,  5,  5,  5,
       6,  6,  6,  7,  7,  7,  7,  8,  8,  8,  9,  9,  9,  10, 10, 10,
       10, 11, 11, 11, 12, 12, 12, 13, 13, 13, 13, 14, 14, 14, 15, 15,