reset column drawing function

Fixes #301

Short analysis: Usually, the colfunc() pointer is reset to
R_DrawColumn() in R_DrawVisSprite() when the player sprites are
rendered via R_DrawPSprite() in R_DrawPlayerSprites() as the last step
in R_DrawMasked().

However, if you bob your 1px wide weapon sprite out of view,
R_DrawPSprite() returns early and the weapon sprite's vissprite is
never rendered, thus the pointer is not reset. That's why the glitch
immediately disappeared when switching to a regular weapon sprite such
as the Chainsaw. This also explains why the glitch never became
apparent in widescreen mode.

Actually, removing the TRANMAP lump or forcing to regenerate it did
*not* fix the issue. It was still there but a lot less visible, since
translucent textures were drawn on translucent textures which were
drawn on the black void - the TRANMAP just helped to make the glitch
undeniable, together with the player sprite that actually triggered
it.

Apparently, the case that I attempted to fix earlier, i.e. that the
TRANMAP lump gets marked as purgeable after being used as a
translucency table for a mid-texture, is already been taken care of
in P_LoadSideDefs2().

Source/r_segs.c

@@ -206,10 +206,11 @@ void R_RenderMaskedSegRange(drawseg_t *ds, int x1, int x2)
         maskedtexturecol[dc_x] = D_MAXINT; // [FG] 32-bit integer math
       }
 
+  // [FG] reset column drawing function
+  colfunc = R_DrawColumn;
+
   // Except for main_tranmap, mark others purgable at this point
   if (curline->linedef->tranlump > 0 && general_translucency)
-   // [FG] make sure to never mark main_tranmap as purgeable
-   if (tranmap != main_tranmap)
     Z_ChangeTag(tranmap, PU_CACHE); // killough 4/11/98
 }