Let super().end_scan() run even if the 2BM optional steps fail

[xmap]

Companion to the fix in fix/end-scan-must-not-wedge-startscan, branched
separately from master so the two can be reviewed and merged
independently. That PR stops a raise inside end_scan() from killing the
scan thread outright. This one removes the most concrete cause of such a
raise at 2-BM: the two unguarded, best-effort steps this override adds
before calling super().end_scan().

Mechanism. TomoScan2BM.end_scan() does two things beyond the base
class's own cleanup: it appends a web camera frame to the scan file with
fid.create_dataset('exchange/web_camera_frame', data=frame), and it
copies the finished file to the analysis computer with dm.fdt_scp() or
dm.scp(). Neither block is guarded. If either raises, super().end_scan()
never runs, so ScanStatus is never set to its final value, StartScan is
never put back to 0, and scan_is_running is never cleared. This is
exactly what happened on 2026-08-20: a full disk truncated the scan file,
create_dataset raised ValueError: Unable to synchronously create dataset (address of object past end of allocation), and the scan thread died
before super().end_scan() ran.

add_theta(), called a few lines earlier in this same method, already
wraps its own hdf5 access in a bare except that logs an error and prints
a traceback, on the same reasoning: a problem writing one auxiliary dataset
should not stop the rest of end_scan() from completing. add_theta()
survived the identical failure on the identical file during the incident.
This gives the webcam-frame block and the data-transfer block the same
guard, each in its own try/except, matching add_theta()'s style and log
level (log.error plus traceback.print_exc(file=sys.stdout)).

Scope. Only tomoscan/tomoscan_2bm.py, only the end_scan() method.
The two blocks are each wrapped in a try/except; nothing inside either
block changes, and super().end_scan() is unchanged. No other method is
touched. The re-indentation needed to wrap the blocks also strips some
trailing whitespace on lines inside them; that is incidental to adding the
guards, not a separate cleanup pass.

Testing. This repo has no test coverage of end_scan(), and building
an EPICS harness for it is out of scope for this change. To reproduce:
make either block raise (for example, point the credentials file at a
missing path, or force create_dataset to fail against a truncated file)
and confirm super().end_scan() still runs and StartScan reaches 0.