fix(sbgecom): fix stack overflow when sending frames on NuttX, fix serial port misconfiguration (B0 in SITL), fix timestamps and validity flags on mag data in both directions

[tolesam]

Solved Problem

The sbgECom driver had several defects breaking both directions of the link:

• Every transmitted frame overflowed the TTY work queue stack on NuttX (the
  sbgECom library puts a 4096-byte buffer on a 1728-byte stack), corrupting
  heap memory.
• A broken termios override (c_cflag &= CSIZE) cleared CREAD and set the
  port to B0 in SITL.
• Transmitted mag aiding logs carried status = 0 (failing BIT, out of
  range, uncalibrated) and a bogus timestamp, so the INS rejected them.
• Received mag data was published with the INS timestamp instead of PX4 hrt
  time.

Solution

• Run the driver on the INS0 work queue (6000-byte stack).
• Remove the termios block — sbgInterfaceSerialCreate() already configures
  raw 8N1.
• Outgoing mag logs: set BIT/in-range/calibration-ok flags, send timestamp 0
  and let the INS stamp on arrival.
• Incoming mag data: publish with hrt_absolute_time(), like accel/gyro.
• Cleanup: cancel perf counters on error paths, drop duplicate send-failure
  warnings.

Changelog Entry

For release notes:

Bugfix: sbgECom driver - fix stack overflow when sending frames on NuttX, fix serial port misconfiguration (B0 in SITL), fix timestamps and validity flags on mag data in both directions
No new parameters.
No documentation changes.

Alternatives

N/A

Test coverage

• SITL with a recorded sbgECom stream replayed over a pty: received topics
  publish with coherent PX4 timestamps; outgoing mag frames captured and
  checked for status flags and zero timestamp.
• Bench test against an SBG INS on serial: no more send-side reception
  errors; INS accepts the mag aiding logs.

Context

N/A