Follow-ups to lz4 decompression bug

[embray]

PR #193 was done hastily so that @braingram could have it fixed for a demo tomorrow.

Should do some follow-ups on this though:

• Clean up handling of the LZ4 block header read state; currently rather ad-hoc in asdf_compresser_lz4_read_block. Thankfully in the doc comments for the function asdf_compressor_lz4_read_header I wrote "reset the position to 0 to re-read from the current input position" which helped me find the bug quickly, because my own code forgot to do that. But this should be improved.
• Add a regression test -- it seems that none of the test files for compression are large enough that there would be more than one LZ4 compression blocks.
• I also found an unrelated bug (introduced by Add asdf dd command to the CLI #192) in asdf_ndarray_read_tile_ndim where a call to asdf_ndarray_data_raw wasn't updated to asdf_ndarray_data. Thought I had gotten all of them. But there should be a regression test for this as well (I don't think any of the tests call asdf_ndarray_read_tile_ndim on a compressed array).