fix(gate-61): an empty scope has two causes and only one of them is a diff (.github#347)

[rubenvdlinde]

Closes #347.

bin/hydra-gates forwards --base only on a diff-scoped run, so on --full the
runner keeps its own origin/development default and hands it over anyway. On
development itself that diffs the branch against itself, the checker returns 3,
and gate-61 printed

[gate-61] listener-work-placement: NOT APPLICABLE — the diff against
'origin/development' put every post-event registration out of scope

two lines after the run's own preamble said Base ref: n/a — --full requested.
There was no diff. Sibling of #361: check_listener_placement.py:555 carries
the same non-empty argparse default.

The verdict does not change, and that is deliberate

Sweeping the tree on an unscoped run was tried here before and reverted — the
runner's own invocation comment records it. The builder runs unscoped, so
--all surfaces the fleet's whole registration backlog as blocking findings on
every build. na with an honest reason is the third answer: it cannot go
false-RED, and it stops the run claiming an exclusion nothing performed.

Blast radius: zero. No verdict changes in any repo, at any scope.

What was actually missing was the number

0 of 1 inspected and 0 of 45 inspected printed identically. An advisory
whole-tree sweep now supplies it — informational, exit status discarded, and quoted
only when the helper printed its own terminal summary, so a sweep that finds
inherited debt cannot become this run's verdict and a sweep that crashes cannot
either.

Measured on openregister @ development: --all reaches 45 post-event
registrations carrying 3 findings — exactly the number SHARED-LESSONS records as
hidden by this defect. It is now on the verdict line.

⚠️ The advisory summary is recomposed, not quoted

The helper's own line ends …, 0 out of scope: 3 failure(s). Pasting it in put the
phrase out of scope straight back into a reason on a run that computed no
scope — the exact sentence this fix removes, smuggled in as a quotation. The suite's
gate-agnostic property (a NOT APPLICABLE may not blame a diff on a --full
run) caught it, listing gate-61 among the blamers again after the "fix". Worth
recording: the generic invariant caught a regression the gate-specific assertions
would have passed.

Fixture — red before, green after

Same tree, only the scope flag changed:

	result
before	3 assertions RED — "#347 is LIVE: gate-61 still claims 'the diff … put every post-event registration out of scope'"; the reason does not state the run computed no diff; the backlog size is not stated
after	13 passed / 0 failed, and gate-61 drops off the "gates blaming a diff on a --full run" list, which now reads 6 7 8 9

The _known_defect entry for #347 is replaced by positive assertions, as its own
text demanded ("flip this assertion and delete the entry"). The gates 6/7/8/9
wording defect is a separate, still-unfiled entry and is untouched.

ShellCheck clean.