Fix an incorrect check

[res2k]

...introduced in 0f5ef84

An input check condition required some tweaking which was missed when porting a commit.

[apanteleev]

Merged, but can you please explain what's the idea behind negative bits? I see its sign being removed right at the beginning of the function, but some places actually call MSG_WriteBits with negative bits. Why?

[res2k]

I think the answer can be found in MSG_ReadBits(): There, a negative count of bits indicates "sign extension".
Say, you have the bits "0xf000". MSG_ReadBits(16) would return 0x0000f000, MSG_ReadBits(-16) would return 0xfffff000.
The same scheme is used for MSG_WriteBits(). While it doesn't change the results, using negative number communicates that a signed value is being written, so the corresponding MSG_ReadBits() for the data should also use a negative bit count, lest the value may not be interpreted correctly.

[apanteleev]

That makes sense, thanks. It's quite an obscure solution though.

[res2k]

It's simultaneously obscure yet somehow strangely sensible ("You want a signed value? Put a sign in front of it's bit size.")