Add buffer/push-* sized int and float

[pnelson]

I suspect many binary protocols need to encode/decode floats to/from big/little endian. Here is a first draft, should it be included in the standard library.

I am not tied to the binary module as I'm aware there exists buffer/* functions that are similar and it could possibly be bundled into the existing buffer module. I was concerned about the naming because buffer/push-word already exists and the signature is different with no control over endianness. I thought it might be confusing to add buffer/push-uint32 in that case - but maybe not.

Additionally, binary/write-float32 and binary/write-float64 could potentially be removed if we implement something like math/float32-bits and math/float64-bits respectively for constructing sized uints from the underlying bits and then lean on binary/write-uint32 and binary/write-uint64 to write to the buffer.

I just wanted to start these discussions and fix any outstanding bugs that may be present.

[pnelson]

Self-review comments.

[pnelson]

I was thinking this should be janet_getuinteger which exists below janet_getinteger in src/core/capi.c but I was getting these warnings/errors on make valtest:

cc -O2 -g -std=c99 -Wall -Wextra -Isrc/include -Isrc/conf -fvisibility=hidden -fPIC -c build/c/janet.c -o build/janet.o
src/core/binary.c: In function ‘cfun_binary_write_uint16’:
src/core/binary.c:86:25: warning: implicit declaration of function ‘janet_getuinteger’; did you mean ‘janet_getinteger’? [-Wimplicit-function-declaration]
   86 |     u.data = (uint16_t) janet_getuinteger(argv, 2);
      |                         ^~~~~~~~~~~~~~~~~
      |                         janet_getinteger
src/core/capi.c: At top level:
src/core/capi.c:296:10: error: conflicting types for ‘janet_getuinteger’; have ‘uint32_t(const Janet *, int32_t)’ {aka ‘unsigned int(const Janet *, int)’}
  296 | uint32_t janet_getuinteger(const Janet *argv, int32_t n) {
      |          ^~~~~~~~~~~~~~~~~
src/core/binary.c:86:25: note: previous implicit declaration of ‘janet_getuinteger’ with type ‘int()’
   86 |     u.data = (uint16_t) janet_getuinteger(argv, 2);
      |                         ^~~~~~~~~~~~~~~~~
make: *** [Makefile:223: build/janet.o] Error 1

[pnelson]

Maybe there is a historical reason for this but I found that I couldn't write an 8 byte uint64_t. This function checks for the unsigned integer range using JANET_INTMAX_DOUBLE which is 9007199254740992.0 or hex 20000000000000 (7 bytes).

[bakpakin]

This should almost certainly go in the buffer module, but otherwise seems reasonable.

The ffi module and peg module have some of this functionality but with other limitations. I would prefer if this were just folded into the buffer module as "binary" isn't particularly descriptive for a module name.

[pnelson]

int/to-bytes used nil to represent native endianness. I felt :native was more descriptive but I would understand the desire to maintain consistency with what already exists.

[sogaiu]

Another path to consistency might be for int/to-bytes to also support :native?