Support all JOINT formats supported by glTF specs

* Flip release of buffer;

First it needs to be taken out of the processing chain, then it can be released. The inverse of the initialization.

* Add mutex locks to audio buffer functions; Separate those used from both threads

* Flip release of buffer;

First it needs to be taken out of the processing chain, then it can be released. The inverse of the initialization.

* Remove TODO marker; The buffer is in stopped state and its data won't be accessed

* Add mutex locks to music/stream functions directly operating on buffer

* Secure UpdateMusicStream/PlayMusicStream/UpdateAudioStream;

This change is twofold:
* Add locks to UpdateMusicStream/UpdateAudioStream (second one needed separation)
* Remove unnecessary hack to restart music - inlining the statements resulted in a no-op

Especially the second part made it easier to ensure thread-safety overall

* Remove redundant check; Already checked at beginning of function

This could be actually very tricky and GPU/drivers dependant...

For consistency, now _almost_ all `Draw*Lines()` functions use `RL_LINES` mode for drawing. It solves the linked issue but it can have other implications, as mentioned in the WARNING comment in `DrawRectangleLines()`.

Side note: `DrawRectangleRoundedLines()` now should be reviewed for consistency.

- Review formatting
 - Improve readability for some functions result return
 - Minimize early returns
 - Align LoadFileData() to UnloadFileData()

The way the current code worked was by calling `GetCurrentMonitor()`,
which would always return the monitor at position (0,0). This isn't the
primary monitor on all platforms, on Linux in particular it isn't the
case.

This isn't the case on the SDL backend, after calling `InitWindow()` the
window would always show up on the primary monitor. Even on the GLFW
backend, if the full screen flag was set it would attempt to put it on
the primary monitor as it would call `glfwGetPrimaryMonitor()` to do it,
so for consistency's sake we should do it on windowed mode too.

This reverts commit c1fd98591d.

The key handling in PLATFORM_DESKTOP_SDL was faulty in two ways, which
led to GetKeyPressed returning incorrect data.

CORE.Input.Keyboard.keyPressedQueue was updated only on SDL_TEXTINPUT,
meaning only text characters were registered as a pressed key, but not
function keys (eg. tab, backspace...). Also on such event, both
CORE.Input.Keyboard.keyPressedQueue and
CORE.Input.Keyboard.charPressedQueue were assigned the key's
corresponding codepoint, when CORE.Input.Keyboard.keyPressedQueue
should get the raylib keycode instead.

CORE.Input.Keyboard.keyPressedQueue is now updated on SDL_KEYDOWN event
instead.

Co-authored-by: Arthur <hi@thenightwat.ch>

Fixes some crashes, e.g. calling StopMusicStream after trying to load Music from a non-existant .mp3 file

* fix local dependency breakage

* revert accidental change in include path

Line 110 of file rlgl.h is the one in question.

* update to latest zig, and don't use srcdir

* preserve compatibility, simplify Child.run compatibility

Fix typo

add gamepad previous button state tracking to Android

Changed the raylib version from 4.2.0 to 5.1-dev, just like the raylib.h file

The changes brought by #3891 uses `std.process.Child.run` which was
renamed from `std.process.Child.exec` in Zig version 0.11.0. This commit
adds a version check to use the appropriate function names.

Additionally, the `linux_display_backend` configuration option is added
so users can set it when running `zig build` with the `-D` option or
when using raylib as a dependency.