fix(vetting): take up the join from the application it was made for

[stormer78]

Rebased onto main now that #342 has merged; this PR is a single commit against main.

Why

Choosing to be vetted from the join flow abandoned the join. JoinVettingApply
started the application, set active_page = Main and returned Returned(None);
nothing was carried over, so once the statements were in the community had to be
found and entered a second time. That is the whole reason vetting read as
something to go and do before joining, rather than as one of the ways of
joining — it was not a prerequisite, it was a dead end mid-flow.

What changed

j on a vetting application takes up its join. The community is the one the
application names, so Action::StartJoinFor(vtc_did) → JoinEntry::ForCommunity
opens the flow where the DID entry page would have led: same discovery, same
routes chooser, no DID to find again.

Nothing new is persisted. An application already records its community, so it
is the parked join. That is better than a transient handle on the loop would
be: gathering statements takes days, and the person will restart OpenVTC many
times in between. A field that did not survive a restart would be a resume that
works only in the session where it is least needed.

The submit arm's body moves into enter_community. Both ways into a
community — the entry page and a resumed join — now share one path, so the
invitation collection, the manifest question and the "could not ask" case cannot
drift between them.

The degraded loop's arm takes both verbs. StartJoin | StartJoinFor(_) build
the entry and fall into the same body, which keeps the "must never own a live
listener across an iteration" invariant documented over the pair rather than over
one of them.

Two "next step" lines stop pointing elsewhere. next_step_words(Join) was
"join from Communities (j)" and the satisfied progress line was "join from
Communities"; the key now works where the person is standing, so they say so. The
messages the join flow leaves on the Vetting panel say the join is not lost.

Testing

Re-run after the rebase:

• cargo fmt --all --check, cargo clippy --all-targets --all-features — clean
• cargo test --all-features -- --include-ignored — all green
• RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --all-features — clean

New test: j sends StartJoinFor with the community of the row under the cursor
(not the first row), does nothing with no application under the cursor, and is not
bound on the desk.

The manual dry-run guide is updated for both PRs — a new step 06a (start from the
join, read the ways in), a rewritten step 11 (take the join up again, after a
restart, to prove the application is the parked join), and four new "ways in"
negative cases.

Live run against a vetting community not yet done.