feat: add decode_comprehensive fuzz target with safe metadata probing

New fuzz target exercises TiffHeader::parse, dimensions/colortype
probing, chunk-by-chunk reading, and multi-IFD iteration. Guards
get_chunk_type/chunk_dimensions behind dimensions() success check
to avoid documented panics on non-image IFDs.

fuzz/Cargo.toml

@@ -23,3 +23,9 @@ name = "decode_image"
 path = "fuzz_targets/decode_image.rs"
 test = false
 doc = false
+
+[[bin]]
+name = "decode_comprehensive"
+path = "fuzz_targets/decode_comprehensive.rs"
+test = false
+doc = false