Core: Add LV_NODISCARD for C API functions with non-discardable results

[kaixiong]

This PR adds the define LV_NODISCARD for marking C API functions whose return results should not be discarded. I have added them to API functions that return allocated objects.

The define uses compiler-specific function attributes to achieve this when building C code.

For C++, [[nodiscard]] already exists from C++17 onwards. C++20 has a variant that accepts a string literal for explaining the rationale that could be displayed in compiler messages. I can't see how to achieve the same for C unfortunately.

Use LV_NODISCARD only for C API functions. [[nodiscard]] for C++ functions.

[hartwork]

@kaixiong I like the idea. I think I found three cases that slipped through the cracks:

# git grep 'LV_API.*_new' | grep -v LV_NODISCARD
libvisual/libvisual/lv_param_value.h:LV_API VisParamValue *visual_param_value_new        (VisParamType   type, void *value);
libvisual/libvisual/lv_time.h:LV_API VisTimer *visual_timer_new  (void);

# git grep 'LV_API.*clone' | grep -v LV_NODISCARD
libvisual/libvisual/lv_palette.h:LV_API VisPalette *visual_palette_clone (VisPalette *self);

What do you think?

[hartwork]

@kaixiong could it be that this needs addition of defined(__cplusplus) && ?

[kaixiong]

@hartwork, I was wondering about this. GCC and Clang compile it but I'm not sure if this is standard. Any idea?

[hartwork]

@kaixiong let me first spell out the questions that I think we need to answer and then try to give a potential answer. For the questions:

1. Is this code used with a C compiler or only C++? Does need it to work from plain C?
2. Can we safely compare an undefined macro against an integer value?
3. Does the __cplusplus value reflect reality (more below)?

Potential answers:

1. Seems like yes, we need to support both C and C++ here.
2. It is defined according to https://stackoverflow.com/questions/5085392/what-is-the-value-of-an-undefined-constant-used-in-if so it seems to become a question of style and/or clarity.
3. MSVC is said to not have reality in __cplusplus. The post lays out some options. Another one. I think we do want to support MSVC?

PS: https://sourceforge.net/p/predef/wiki/Standards/ does not seem to have any additional insights.

What do you think?

[kaixiong]

@hartwork Looks like Question 1 and 2 are effectively the same question as the C/C++ preprocessor both default to 0 for undefined constants (apart from the handling of true and false).

Question 3 is addressed in #269 but does not ultimately matter for this PR because the code falls back to compiler extensions (_Check_return_) immediately below this.