fix: VLSD signal data is unreadable after a composed channel

[kipp-ing]

Background — what VLSD is

Most MDF4 channels are fixed-width columns inside the record: the channel says
"my value lives at byte offset 12, 2 bytes wide", and every sample is the same
size. Some data cannot work that way — a text channel, or a CAN-FD payload whose
length changes from frame to frame. MDF4 stores those out of line: the record
holds only a 64-bit offset, and the actual bytes sit in separate signal-data
blocks elsewhere in the file. Such a channel is a VLSD channel — variable
length signal data.

To read one, asammdf has to know where that channel's signal-data blocks are.
It collects them per group in Group.signal_data, a plain list indexed by
channel index, filled while the channel list is parsed. If that list ever falls
out of step with Group.channels, a VLSD channel looks up somebody else's blocks
— or none at all.

That is what happens here.

Problem

MDF4._read_channels appends an extra grp.signal_data.append(None) for
every channel that has a component_addr (a structure or array composition),
while dependencies gets exactly one entry per channel.

So every composed channel shifts the signal-data entry of all following channels
by one, and _load_signal_data then looks in the wrong slot. Reading such a
channel raises:

MdfException: Wrong signal data block refence (0x4AD5260) for VLSD channel "CAN_DataFrame.DataBytes"

Impact

This affects every CAN-FD bus-logging measurement recorded with variable-length
payloads — which is the normal case, since a CAN-FD frame carries anywhere from
0 to 64 bytes. CAN_DataFrame.DataBytes is a VLSD channel, and it lives inside
the composed CAN_DataFrame structure, so it is always one of the shifted ones.
MDF.extract_bus_logging() cannot read the payload at all and the file is
effectively unusable.

Found on two real OEM CAN-FD bus logs (1.0 GB and 157 MB, 5.1 M and 2.3 M CAN
frames); with the fix both extract normally.

Fix

Drop the stray append. grp.signal_data then stays in step with grp.channels.

Test

test_vlsd_channel_after_structure_composition in test/test_mdf4.py — writes a
VLSD (string) channel after a structure channel in the same group, then reads it
back. Fully offline, no downloads.

Without the fix:

AssertionError: 6 != 5                                     # len(signal_data) vs len(channels)
MdfException: Wrong signal data block refence (0x7D0) ...   # reading the channel

test_mdf, test_mdf4, test_CAN_bus_logging, test_endianess, test_cantp
and test_signal all pass.