As generation is done in software, it's up to the user to do it. `ImageMipmaps()` is already provided for reference.

Fixes #2683

Remove elements from touch point related arrays when touch up and
similar events are processed. This makes GetTouchPointCount() always
report the actual count of touch points, and all positions returned by
GetTouchPosition() correspond to positions of currently happening
touches.

* [rlgl] Check for extensions before enabling them

* Shift to glad on macOS

* #undef CORE_OPENGL_33

* Remove version hack and fix ASTC compression assumption

* Remove loader from glad

* Use GLAD_MALLOC/FREE instead of malloc/free

* More explicit extension checking

Currently assumes a single gamepad, has no code specific to gamepad
detection (gamepad is "detected" when an event related to gamepad
arrives). Also assumes that all gamepads look roughly like an xbox/ps
controller. Both assumptions are not strictly true, but an
implementation like that probably covers 85% of usecases.

Also it doesn't update previousButtonState so functions
IsGamepadButtonPressed() and IsGamepadButtonReleased() don't work, but
they didn't work previously on Android anyway, and they are flaky on
desktop as they are now, so the mechanism for these two functions
probably should be reworked anyway.

It's certainly an improvement compared to the previous android gamepad
handling code, which put gamepad events into touch related structs.

On android after rebinding context (which happens when you minimize and
navigate back to an app, or when you turn a screen off and back on)
there's a bug that viewport has a wrong scale and part of it is off
screen.

The change fixes it on devices I tried, but the solution feels hacky to
me. However when I attempted instead to call SetupViewport() again which
feels like a more proper solution, it didn't fix it.

Fix display -> screen coordinate conversion for android platform and
move it to the platform event handling code, simplifying
GetTouchPosition() function implementation.

Co-authored-by: Denis Pobedrya <denis.pobedrya@gmail.com>

raylib library tries to avoid `long long` usage. Several SSBO functions have been reviewed (including some renames for consistency) to minimize `long long` type usage.

This simplification will allow the usage of `rshapes` as STANDALONE mode in a future. Only a small set of `rlgl` functions are required and they can be "more" easely replaced if no `rlPushMatrix()`/`rlPopMatrix()` are involved.

More simplification planned for the future, maybe the textures dependencies.

Updated version to `rlgl 4.2`

* Update raylib.h

* CheckCollisionPointPolygon()

* typo

I want to note this is a special symbol exported

Reasons to NOT define `_GNU_SOURCE`:

 - access to lots of nonstandard GNU/Linux extension functions
 - access to traditional functions which were omitted from the POSIX standard (often for good reason, such as being replaced with better alternatives, or being tied to particular legacy implementations)
 - access to low-level functions that cannot be portable, but that you sometimes need for implementing system utilities like mount, ifconfig, etc.
 - broken behavior for lots of POSIX-specified functions, where the GNU folks disagreed with the standards committee on how the functions should behave and decided to do their own thing.

WARNING: This could be a BREAKING CHANGE for some platforms! I'm afraid something could be wrong on `rglfw.c` module.

To be able to compile on Windows I had to modify `glfw/src/platform.c` line 74. I couldn't manage to compile without that change, help is welcome!

This linkage is only required for arm 32bit but I don't know how to detect that specific architecture in the Makefile...

* Win32: resolve some symbols re-definition of windows.h in glfw3native.h

This reflects GLFW's fix: https://github.com/glfw/glfw/issues/1348

This enables to build with a external GLFW containing the
following fix:

* 05f6c13d11

Currently, glfw3native.h of the internal GLFW is customized at
2feea87b61

This fix is compatible with the current customized glfw3native.h.

This fix enables us to update it to the latest and remove the
customization.

* Win32: remove unneeded typedef

When file extension is longer or equal length compared to buffer holding lowercased string, strncpy does not null terminate the string.
Increased buffer size by 1 to ensure it will always be null-terminated, so that following strcmp does not read out of buffer bounds.

This truncation causes text that spans the zero coord boundary to round differently left or zero vs. right of zero, in turn causing letters to appear squished together. If you actually need the position to be an integer, you should instead `floorf()` the float, rather than doing an integer truncation, but I don't see any reason to convert it to an integer in the first place. Everything else in the equation is a float.