Rule out a bit-banged board without writing to it

[Boernsman]

Follow-up to #2, which claimed this was already handled. It was not.

What was wrong

#2 made a bit-banged board give itself away by handing over bytes with nothing
asked of it, and skipped the write when it did. That check passed in testing only
because a libftdi session had just run against the board — which is what wakes the
read side of an FT245's FIFO. On a fresh port the board stays silent for as long as
it is listened to (measured: nothing in 2 s), so the probe went ahead and wrote.

The result on main today, with the relays at 0x5a:

$ denkovi status
type16 DAE00745 on /dev/cu.usbserial-DAE00745, 16 relays
...
on: 3, 5-8, 10-11, 16  [0x86f4]

ask// landed on the relays and left them holding a / (0x2f), and the bytes the
FIFO handed back were taken for a status reply, so an 8 relay board was reported as a
type16 with sixteen relays that are not there.

The fix

Read the chip's data lines before anything else. read_data_lines opens the chip in
its ordinary serial mode and only reads: no byte is sent, no pin is switched to an
output, so it is safe to do to a board of any kind. But it wakes the FIFO, so a
bit-banged board then answers the port unprompted and is recognised before a single
byte is written to it.

Measured across three patterns — 0x5a, 0xa5, 0xf0 — the auto-probe now leaves
the relays exactly where they were and asks for --board type8 instead of inventing
a type16.

probe_board_type and resolve_board_type take the Device rather than the port
name, since the serial number is what finds the chip.

When libftdi is not there

Nothing can be told about the data lines, so the board falls through to the protocol
probe exactly as before. The worst that comes of it is a board that has to be named
with --board, which is already the documented answer for these boards. There is a
test for that path.

Testing

Against the 8 relay board (FT245) on macOS 26: the three-pattern probe check above,
and on, toggle, off still driving the board. 80 tests pass; the probe tests
stub both the FTDI chip and the serial port, including the no-libftdi fallback.