In NDR each digital_in channel represents a single bit. Previously
the SpikeGLX reader exposed the entire packed digital word as a
single 'di1' channel, which conflicted with the NDR convention and
made it impossible for callers to ask for an individual line.

Detect digital width from metadata:
  - NIDQ: n_digital_lines = 8 * (niXDBytes1 + niXDBytes2). For the
    user-reported file (niXDBytes1=1) this gives 8 lines (di1..di8).
    A future file with niXDBytes1=2 would yield 16, and adding port1
    via niXDBytes2 extends further. Falls back to 16 * n_dw_chans
    when the byte fields are absent.
  - IMEC: n_digital_lines = 16 * n_sync_chans, so the int16 sync
    word is exposed as di1..di16 (bit 6 is the SMA sync input in
    practice; other bits are available if a recording uses them).

header.m gains two new fields, n_digital_word_cols (number of int16
columns in the file holding digital data, always at the end) and
n_digital_lines (number of single-bit lines exposed). Existing
fields are unchanged so samples2volts and the NIDQ tests keep
working.

getchannelsepoch now emits one digital_in entry per bit
(di1..di_n_digital_lines) instead of a single di1.

readchannels_epochsamples('digital_in', channels, ...) maps each
1-based line index to (DW column, bit position), reads each unique
DW column once via read_samples, and extracts the requested bit
with bitget. Returns int16 0/1 values, one column per requested
line. Out-of-range lines raise
ndr:reader:neuropixelsGLX:DigitalLineOutOfRange.

Tests:
  - +ndr/+test/+reader/+neuropixelsGLX/test.m: writes a counter
    pattern to the IMEC sync column and verifies di1, di8 and di12
    return the correct bit via the high-level r.read() API. Adds a
    NIDQ section that mirrors the user's snsMnMaXaDw=0,0,8,1 +
    niXDBytes1=1 configuration, asserts exactly 8 di channels are
    listed, reads di1 via r.read() (the exact failing call from the
    bug report), and verifies di9 raises DigitalLineOutOfRange.
  - tools/tests/.../TestNeuropixelsGLX.m: testGetChannelsEpoch
    updated to expect 16 digital lines after the neural channels
    instead of a single di1.

The neuropixelsGLX header now stores explicit per-line metadata for
the digital channels it exposes:
  - n_digital_lines     : count of active single-bit lines
  - digital_line_col    : 0-based DW column offset for each line
  - digital_line_bit    : 0-based bit position within that column
  - digital_line_label  : underlying SpikeGLX line label, e.g.
                          'XD0'..'XD7' for port-0, 'XD1.<k>' for
                          port-1, 'SY<col>.<bit>' for IMEC sync.

NIDQ line counts come from niXDBytes1/niXDBytes2 (8 active lines per
saved byte). NI hardware only enables digital input in whole-byte
chunks, so every bit within a captured byte is electrically active
even when the user only physically wired some of them. niXDChans1/2
remains informational and is not used to gate which lines are
exposed. IMEC streams have no per-bit configuration, so all 16 bits
of each sync int16 column are exposed.

readchannels_epochsamples no longer recomputes (col, bit) on the
fly; it just looks up info.digital_line_col / digital_line_bit by
the requested 1-based di index. The bit-extraction loop, the unique
column grouping, and the read_samples call sites are unchanged, so
behavior for the user's file (8 di channels mapping to bits 0..7 of
the only DW column) is identical.