bugfix: handle negative values according to their bit size in decode_truncated_block_i32

[mwicat]

Hi,

I tried to use ncw-convert to convert a sample library consisting of 124 samples from several keygroups. Two of those samples caused some errors which I decided to analyze. This is one of them:

$ ncw-convert testfile-truncated-compression.ncw testfile-truncated-compression.wav
Error: TooWide

Looking at the file, it consists mostly of "delta decode" blocks but there is also one "bit truncation" block near the beginning that is not properly decoded by ncw-convert. Looking at the code, sequences of bytes for "bit truncation" scheme are treated like unsigned types which result in numbers that are too big to be encoded as signed type when writing to wave file and that in turn causes value checks in WavWriter::write_sample to throw an error. For example:

• inside 24-bit size "bit truncation" block, a byte sequence [0xfc, 0x31, 0xda] is 24-bit little endian representation of -2477572 but decode_truncated_block_i32 will interpret is as 14299644 which is neither real sample value nor a value that can fit inside signed 24-bit type.

After applying my fix the file seems to be decoded properly, I hope this is in line with how "bit truncation" should work in general. Attaching the file that causes the mentioned error.

testfile-truncated-compression.zip

[cheesymoo]

How do you reconcile this change with #5 ?