Phoenix reader update

[bvkay]

Fixes #7

1. Prefer EMpower's receiver metadata when present: empower_recmeta.json holds corrections made after the survey, so it is taken over recmeta.json when both exist.

2. Join sequence reads once: read_sequence grew the array with np.append per file, copying everything already read each time. Chunks now concatenate once at the end.

3. Fix the 24-bit decode (the Phoenix NativeReader.read() decodes 24-bit samples with a byte from the next sample #7 fix): read() strided 4-byte windows over the 3-byte samples, so the low byte of every decoded value was the next sample's high byte (error bounded at 255 counts, ~5.9e-7 V at the 5 V range) and the last window read one byte past the buffer. Each sample is now padded to 4 bytes with a zero low byte, the same convention as read_frames. Affects the .bin native path only; the float td_* formats are untouched. This changes output values for .bin reads.

[kujaku11]

@bvkay Thanks for the continuous updates. This is great.