HPyBytes_AsString should return const char*

[antocuni]

Currently we have:

char* HPyBytes_AsString(HPyContext *ctx, HPy h);
char* HPyBytes_AS_STRING(HPyContext *ctx, HPy h);

these two functions were quickly added by me because I needed them for ujson. They return char * to match CPython's signature, but I think they should return const char* instead.
The reason why CPython's returns char* is this note in the docs:

[...] The data must not be modified in any way, unless the object was just created using PyBytes_FromStringAndSize(NULL, size)

But I don't think we want to support this use case. Instead, we should introduce StringBuilders (see #181 ).

Moreover, this will be consistent with e.g. [H]PyUnicode_AsUTF8: it used to return char*, but since CPython 3.7 it returns const char*

Another question is whether we want to keep the AS_STRING version at all, and all the similar UPPERCASE macros. Ideally not, but on CPython they might be slightly faster, so we need proper benchmarks to make an proper decision.

[hodgestar]

We should also document how long the returned memory is valid (perhaps only as long as the handle?) and add checks to debug mode that it is not written to and that it is not accessed after the handle is closed (or whatever lifetime we pick).

[fangerer]

@hodgestar: Agreed. We already defined that and implemented exactly that check for HPyUnicode_AsUTF8AndSize (see function

hpy/hpy/debug/src/debug_ctx.c

Line 122 in 8b33bd0

const char *debug_ctx_Unicode_AsUTF8AndSize(HPyContext *dctx, DHPy h, HPy_ssize_t *size)

). The infrastructure to do so is already there, so it shouldn't a problem to do the same for HPyBytes_AsString.