UoA reader updates

[bvkay]

Fixes 05

Updates with:

1. Reader Registration: for Lemi-423, EDL PR6-24 and legacy orange box.

2. Filter Gain Direction: gains are stored counts to physical in pr624.py and lemi423.py. MTH5 divides by the total gain when removing a response, so they must be stored physical to recorded.

3. Filters Marked Not Applied: pr624.py:760 writes applied=False. Aurora only divides out stages flagged applied, so the sensor and interface responses stay in the data.

4. AppliedFilter Field Name: the same call passes sequence=sequence. AppliedFilter takes stage, not sequence, so every stage was written as None.

5. LEMI-120 Sensitivity: updated to 400 mV/nT.

6. LEMI-423 Reader Cannot Run: four errors before it returns, with MTime called positionally at three sites, data_type = "MTBB" where the enum takes BBMT, ch_metadata.filter.name from the pre-1.0 mt_metadata API, and ChannelTS and RunTS used but never imported.

7. Orange Box Format: records are 21 bytes, not 18; the first header field is a sample count, not a sample rate; and the rate comes from the filter point as 10e6 / (512 * filterpoint).

8. AppleDouble Sidecars: Collection.get_files matches ._name sidecars, which share the data extension but hold resource fork bytes.

9. miniSEED PR6-24: added miniseed support for EDL PR6-24, as AusLAMP Victoria was recorded in this format.

10. B423 Collection Summaries: building a Lemi-423 collection read every file in full. read_summary now memmaps the time columns and takes min/max, the same start/end the old sorted parse reported, so summaries come from the 1024-byte header plus two columns.

11. LEMI-424 Speed and the 16-Column Variant: read_metadata tail-seeks instead of parsing the whole file, position columns map unique values only, and multi-file reads join once; the n_samples estimate divides by the first line's width instead of a hard-coded 152. Instruments with gps/clock problems write a 16-column file, time + fields + battery, no GPS block; which the reader could not open at all. The reader now sniffs the first line's field count; position properties return None when there is no GPS block.

12. Repeated Parsing in Phoenix and miniSEED: recmeta.json was re-parsed for every channel reader; now cached per (path, mtime). EDL miniSEED reads pass format="MSEED" so obspy skips its format-detection cascade.

[kujaku11]

@bvkay This is great, thanks for these updates.