[rcore] Document that GetFPS() must be called every frame

[burinc]

GetFPS() is a sampler rather than a query, and the header does not say so.

Each call takes at most one sample, gated on FPS_STEP, writes
GetFrameTime()/FPS_CAPTURE_FRAMES_COUNT into a 30-slot ring, and returns
1/sum-of-ring:

if ((GetTime() - last) > FPS_STEP)
{
    last = (float)GetTime();
    index = (index + 1)%FPS_CAPTURE_FRAMES_COUNT;
    average -= history[index];
    history[index] = fpsFrame/FPS_CAPTURE_FRAMES_COUNT;
    average += history[index];
}

fps = (int)roundf(1.0f/average);

So the ring needs 30 calls before average is the mean frame time. Calling
every frame reaches that in half a second. Calling from a timer, or from a
periodic summary, never reaches it: after n calls only n slots hold anything, so
it returns 1/(n * frame_time/30).

Measured

An app with a steady 17.02 ms frame time, one binary, the call cadence being the
only variable:

cadence	readings
one call per second, from a cold ring	1773, 887, 591, 444, 354, 295, 253, 221
every frame, same process moments later	59, 59, 59, 58, 59

The first row is 1773/n to within a percent at every point, which is the ring
filling one slot per call.

Called every frame, as DrawFPS() does and as the examples do, it is correct.
The reason this is worth a comment is that misuse has no tell: no error, and a
plausible-looking number rather than an obviously wrong one. It took a
frame-time series that stayed flat while the reported FPS decayed to notice.

The change

Comment only, no behaviour change:

-RLAPI int GetFPS(void);                                 // Get current FPS
+RLAPI int GetFPS(void);                                 // Get current FPS (averages the last 30 calls, so call it every frame)

The rlparser outputs carry the header comment as the function description,
so they are updated to match rather than regenerated. That keeps the diff to one
line per file; happy to regenerate them properly instead if you prefer.

Verified against master at 9b2efc45, which is where this branches from.
GetFPS() is byte-identical there and at 6.0. raylib still builds with the
change (comment only).

Entirely reasonable to close this if per-frame calling is considered obvious
enough not to document.

[raysan5]

Thanks but I prefer to keep the comment simple, I think it's not a big deal.