flags and hacks to get MSAN bot going

This disables a few tests in DM:
   - one BlurLargeImage GM maybe is really broken
   - FontMgrAndroidParser uses libexpat, which I've not (yet?) built from source,
     so MSAN can't see into it.

This extends some of the MSAN stifling we added around SkImageDecoder_libjpeg to SkCodec, and skips .wbmps, .pngs, and .bmps.  We're only seeing issues in colortables for .png and .bmp.

I think I can probably back out disabling Codec and the RAW image decodes...
they should all be covered by the libjpeg stifles.

BUG=skia:4550,skia:4900
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1673663002

CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot,Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Debug-MSAN-Trybot

TBR=msarett@google.com

Review URL: https://codereview.chromium.org/1673663002
diff --git a/src/images/SkImageDecoder_libjpeg.cpp b/src/images/SkImageDecoder_libjpeg.cpp
index 6a3ae87..89bfefc 100644
--- a/src/images/SkImageDecoder_libjpeg.cpp
+++ b/src/images/SkImageDecoder_libjpeg.cpp
@@ -509,6 +509,8 @@
     for (int y = 0;; y++) {
         JSAMPLE* rowptr = (JSAMPLE*)srcRow;
         int row_count = jpeg_read_scanlines(&cinfo, &rowptr, 1);
+        sk_msan_mark_initialized(srcRow, srcRow + cinfo.output_width * srcBytesPerPixel,
+                                 "skbug.com/4550");
         if (0 == row_count) {
             // if row_count == 0, then we didn't get a scanline,
             // so return early.  We will return a partial image.
@@ -525,8 +527,6 @@
             convert_CMYK_to_RGB(srcRow, cinfo.output_width);
         }
 
-        sk_msan_mark_initialized(srcRow, srcRow + cinfo.output_width * srcBytesPerPixel,
-                                 "skbug.com/4550");
 
         sampler.next(srcRow);
         if (bm->height() - 1 == y) {