Fixes for #90: beam-mode NameError, suspender guard, CI collection, alignment tests

[sligara7]

Fixes for NSLS2#90, as a PR into this branch so they land in NSLS2#90 before it merges to main. One commit per issue:

• dclm: import forward_watcher_updates — DCLM.set called it without importing it, so every white/mono beam-mode change (and the new change_beam_mode plan) raised NameError. Both test_monochromator_beam_mode_switch cases failed.
• Profile: restore the ops-mode guard on the beam-current suspender — the merge with main replaced the guard from a8feb5c with an unconditional SuspendFloor(beam_current, 100, resume_thresh=390). That would pause plans during MAINTENANCE/SHUTDOWN until the ring is above 390 mA. Restored main's version unchanged.
• pytest: don't collect src/hextools/gui — --doctest-modules imported the GUI modules in CI, where qtpy and bluesky-widgets aren't installed: 13 collection errors on Python 3.11–3.13.
• Tests: call tomo_alignment_scan with its new keyword arguments — the shutters and rotation motor became optional keyword arguments after exposure_time, but the tests still passed them positionally.

How this was verified

Mock tests only (ophyd-async mock devices; no simulated beamline, TST or real beamline). Same environment for both runs: hextools py312 pixi env, pytest -q, run the way CI runs it.

	failed	passed	errors
main (0bc0fe2)	13	115	0
NSLS2#90 head (6a9e06e)	15	112	1 + 13 GUI collection errors in CI
this branch	13	115	0

The 13 remaining failures are the same tests that fail on main (GeRM, Phantom, change_energy, tomography alignment, radiograph). The alignment tests now fail only on main's existing ValueError: True is not a valid ShutterStatus, not on the call signature. The suspender guard is in the profile, which pytest doesn't collect, so it has no automated test here.

Not in this PR

• RE_METADATA_KEYS is undefined in gui/__main__.py:195. It's inside QtRunEngineManagerMonitor, which nothing instantiates, so it doesn't crash; define it or drop the class.
• Type check and lint will stay red: the type checker can't resolve the Qt/bluesky-widgets imports without the gui extra, and lint debt exists in files this PR doesn't touch.
• Design points for discussion are in a comment on Acquisition GUI based on queue-monitor, that also allows for running locally. NSLS2/hextools#90.