lavc/pnm_parser: disable parsing for text based PNMs

P1, P2, and P3 are respectively the text versions of PBM, PGM and PPM
files.

We can not obtain the buffer size using av_imgage_get_buffer_size() as
every pixel in the picture will occupy a random size between 16 and 32
bits ("4 " and "231 " are such example).

Ideally, we could look for the next header (or EOF) in the bytestream,
but this commit is meant to fix a decoding regression introduced by
48ac4532d437790e56b8ed5d0d467dc88685f035.

Fix Ticket #5670

libavcodec/pnm_parser.c

@@ -66,6 +66,8 @@ retry:
         }
 #endif
         next = END_NOT_FOUND;
+    } else if (pnmctx.type < 4) {
+        next = END_NOT_FOUND;
     } else {
         next = pnmctx.bytestream - pnmctx.bytestream_start
                + av_image_get_buffer_size(avctx->pix_fmt, avctx->width, avctx->height, 1);