Missing 16-bit byte-swap definitions in generic endianness fallbacks

[protz]

The generic big-endian fallback (line 100) and generic little-endian fallback (line 128) define 32-bit and 64-bit byte-swap macros but omit all 16-bit variants. The macros load16_le, store16_le, load16_be, store16_be (lines 192-195) use le16toh/htole16/be16toh/htobe16 which will be undefined on these platforms.

Affected: Any non-Linux, non-macOS, non-BSD, non-Solaris, non-Windows platform using __BYTE_ORDER__ (e.g. bare-metal ARM with GCC, some RTOS environments, Emscripten on big-endian targets).

Test snippet:

// Compile with:
//   gcc -D__BYTE_ORDER__=__ORDER_BIG_ENDIAN__ -U__linux__ -U__APPLE__ \
//       -U__FreeBSD__ -U__sun__ -U_MSC_VER -c test.c
#include "krml/lowstar_endianness.h"
int main(void) {
    uint8_t buf[2] = {0x01, 0x02};
    uint16_t val = load16_le(buf);  // ERROR: 'le16toh' undeclared
    return 0;
}

Suggested fix -- add to the big-endian fallback block (after line 100):

#  define htobe16(x) (x)
#  define be16toh(x) (x)
#  define htole16(x)                                                           \
    ((uint16_t)((((x) >> 8) & 0x00FF) | (((x) << 8) & 0xFF00)))
#  define le16toh(x) (htole16((x)))

And to the little-endian fallback block (after line 128):

#  define htole16(x) (x)
#  define le16toh(x) (x)
#  define htobe16(x)                                                           \
    ((uint16_t)((((x) >> 8) & 0x00FF) | (((x) << 8) & 0xFF00)))
#  define be16toh(x) (htobe16((x)))

This issue was found by Claude (Opus 4.6)