I don't know either why it disappeared in the first place.

Worth doing now that that's been merged into next. Also pulls in a
little bit of master.

This commit brings back the old code for W_CheckNumForNamePwad, using memcmp
instead of strncmp, which is readily inlined by modern C compilers and has
more acceptable performance in P_AddWadFile. Also removes some redundant
strlens, saving the result of one call instead.

This commit increases this branch's distance from SRB2 2.2, per Ashnal
and SteelT's note that I shouldn't worry too much about it, especially
when there's performance on the line.

Thanks to Callmore's insistence on me sweeping through functions that
return -1 on error.

Don't sleep w/ fpscap 0 and vsync enabled

See merge request KartKrew/Kart-Public!294

Per @TehRealSalt's recommendation.

fix portal interp view positioning and camera reset

Closes #14 and #12

See merge request KartKrew/Kart-Public!293

Various fixes for 1.6

Closes #17, #15, and #11

See merge request KartKrew/Kart-Public!292

This time without spare slots, because I don't see shenanigans like
WAD_MUSIC happening with skin indices. Verified to work by saving a
replay with 139 skins as the 133th skin and playing it back a few times.
Shouldn't cause issues with netcode either because there are no SINT8s
handling skin indices.

Can be increased up to 255 safely, just not 256. 5 slots left for
shenanigans like Moe Mansion's WAD_MUSIC.

Update gfx.kart hash

CONS_Printf calls CON_Drawer during startup, thread
collision stuff.

Fixes some wild behavior when a lot of hole punching
requests are coming in.

Now that 1.4 is out, maybe this will come in handy in 1.5, should 1.5 be
released. If not, I guess Ring Racers will make this branch redundant
once that comes out.

This reverts commit 5ec44ba3, reversing
changes made to 4c5d69b7.

Doesn't prevent it from happening later if the rules weren't successfully fetched, though.

Implements W_VerifyPK3, thus allowing music PK3s - even those created
with an archiver, thanks to the preceding commit - to be loaded as
"unimportant" files, like music WADs can be.

Notably caused by a difference in behaviour between Software and GL that was already resynchronised in the 2.2 codebase.

Backports the fixed ResGetLumpsZip code, registration of longname
fields and support for PK3s without trailing directory entries.

Seems to be a regression in internal too. Requires more analysis on behaviour in the 2.2 engine, but essential for maintaining compatibility in 2.1 (such as with Opulence wobbly carpet).

- `cv_masterserver` - Replaces a hack checking for the original SRB2 MS addres every time, and additionally catches those who tested an early version of the http MS.
- `cv_usegamma` - This was previously WAY too easy to accidentially change. This way, a user who has had this accidentially happen will automatically return to baseline.

I originally wrote a ridiculous attempt at reimplementing the existing value rollaround. For one reason or another, that sidesteps net transmission. Let's just avoid all those hacks and limit our influence to changing the maximum allowed range.

Doesn't affect the conditions that allow entry -- just avoids pointlessly resetting the map if it occurs during the period where literally nothing can happen.