fix missing f prefix in the trigger keep_columns comprehension

[artlbv]

Fixes #155.

The bug

hbw/config/config_run2.py:1397 builds the per-trigger HLT entries of
keep_columns["cf.ReduceEvents"] with a set comprehension that is not an f-string:

} | {
    "HLT.{trg.hlt_field}" for trg in cfg.get_aux("triggers", [])
}

So it adds the literal string "HLT.{trg.hlt_field}" exactly once, rather than one entry per
registered Trigger. One character.

Impact — smaller than I first claimed in #155

I originally wrote that this broke what ComputeTriggerSF asks for. That was wrong, and I want
to correct it here rather than leave it in the issue:

• hbw_trigger_selection's init (hbw/selection/trigger.py:188-192) adds the same set,
  correctly f-stringed, to keep_columns["cf.ReduceEvents"]. Any selector routing through it
  already keeps its registered triggers' HLT columns.
• Downstream code reads trigger_ids — built in the selector at trigger.py:52-134, kept by
  name at config_run2.py:1388 — not the HLT path columns. require_triggers,
  masks_dih.py, and hbw/production/trigger.py are all id-based.

So this is latent, not active, and the DL analysis is unaffected. Thanks to @mafrahm for the
trigger_ids point, which is what prompted the recheck.

It bites in exactly one place: a selector that does not route through hbw_trigger_selection
and still wants a registered trigger's raw HLT column. That is the SL channel, which uses the
year ladder in sl_remastered.py — I hit it needing
HLT.Mu12_IsoVVL_PFHT150_PNetBTag0p53 to survive ReduceEvents for a trigger study.

Why fix it anyway

The line does not do what it reads as doing, and the working duplicate in trigger.py is
labelled # testing:, so it is not obviously the load-bearing one. Leaving a silently inert
comprehension next to a hand-maintained list of seven HLT paths is a trap for whoever next
assumes registering a Trigger is sufficient to keep its column — which is what the code
appears to promise.

Worth a separate look (not in this PR): the two mechanisms are redundant, and the # testing:
block in trigger.py is arguably the one that should go once this one works.

Change

The f prefix, plus a comment recording that the effect was masked so the next reader does not
have to re-derive it. No behaviour change for DL: the columns it adds are already added by
trigger.py.

tests/run_linting clean.

[artlbv]

@mafrahm fyi