test-lnk_wsg_resolve live tests route to wrong DB (skip-gate vs lnk_db_conn mismatch)

[NewGraphEnvironment]

Problem

The two live-DB tests in tests/testthat/test-lnk_wsg_resolve.R (closure
mode, line ~138; province mode, line ~148) fail with:

Error: Failed to prepare query : ERROR:  relation "public.wsg_outlet" does not exist

even when a valid database is available. This is a test-harness
connection mismatch, not a real missing-table problem — the skip gate
and the query run against two different databases.

Root cause

• skip_if_no_db() (tests/testthat/setup.R:19-45) opens + validates a
  connection to the local Docker fwapg (127.0.0.1:5432, db fwapg),
  which has public.wsg_outlet (246 rows). Because that connection
  succeeds and is writable, the test does not skip. The helper returns
  that conn (with a deferred disconnect).
• The two closure/province tests ignore the returned conn and call
  lnk_wsg_resolve(cfg, loaded, ...) with no conn argument.
• lnk_wsg_resolve (R/lnk_wsg_resolve.R:122-124) then opens its own
  connection via lnk_db_conn(), which (via PG_*_SHARE env vars / the
  SSH tunnel) lands on the remote bcfishpass database. That DB does
  not have public.wsg_outlet → the query errors.

So the gate greenlights on local fwapg (table present) while the actual
query dials the remote bcfishpass DB (table absent). public.wsg_outlet
is correct and present where it belongs (fwapg); the tests just route to
the wrong DB.

public.wsg_outlet is a 246-row FWA watershed-group outlet-topology table
(one row per WSG, outlet as an ltree) consumed by
fresh::frs_wsg_drainage() for downstream drainage-closure ordering — the
backbone of lnk_wsg_resolve() closure mode and DS-first study-area runs.

Fix options

1. Thread the gated conn through (preferred, smallest): capture the
   conn from skip_if_no_db() and pass it explicitly:

   conn <- skip_if_no_db()
   expect_identical(
     lnk_wsg_resolve(cfg, loaded, wsgs = c("PARS", "BULK"), conn = conn),
     expected
   )

   This makes the gate and the query use the same DB by construction.
2. Point the resolver's default connection at fwapg, or make
   skip_if_no_db() additionally assert public.wsg_outlet exists on the
   connection the resolver will actually use. More involved; only needed if
   lnk_db_conn()'s default target is intended to be fwapg.

Option 1 is the surgical fix and mirrors how the gated conn is meant to be
consumed.

Scope / notes

• Pre-existing; surfaced while running the full suite during link#221
  (accessible_km) work — unrelated to that change.
• Two other closure/strict tests that use .wsg_stub() (offline mocks) are
  unaffected; only the two live-DB tests hit this.