Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@tests/tap/t/018_upgrade_schema_match.pl`:
- Around line 150-165: psql_query currently writes a temp file and may leave it
behind if the function dies before the final unlink; modify psql_query to
guarantee cleanup by using a safe temporary-file helper (e.g., File::Temp->new
or a block-scoped temp file) or wrap the write/psql invocation in an
eval/finally-like construct (local $@; eval { ... }; my $err = $@; unlink
$tmpfile if -e $tmpfile; die $err if $err) so that the temp file is always
removed; reference the psql_query function and the $tmpfile variable when making
the change.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@sql/spock--5.0.6--5.0.7.sql`:
- Line 37: Replace the invalid PL/pgSQL assignment-from-query "target_id :=
node_id FROM spock.node_info();" with a proper SELECT ... INTO form; e.g. use
"SELECT node_id INTO target_id FROM spock.node_info();" (or "SELECT INTO STRICT"
if you expect exactly one row) for both occurrences referencing target_id,
node_id and spock.node_info() (the instances around lines 37 and 109).

In `@tests/tap/t/018_upgrade_schema_match.pl`:
- Around line 276-280: The stop_pg helper ignores failures from pg_ctl which can
leave the old postmaster running; modify stop_pg (the function `stop_pg`) to
check the result/exit status of `run_logged("$PG_BIN/pg_ctl", 'stop', ...)` and
fail the test (or die) if pg_ctl returns non-zero, and optionally wait/retry
until the postmaster is confirmed stopped (e.g. poll for PID file removal or use
`pg_ctl status`) before returning so subsequent phase 6 upgrade assertions won't
run against a still-running old postmaster.
- Around line 156-161: The current code reads psql output from a piped
filehandle ($fh) but never checks the exit status of the spawned psql process,
which can return non-zero and produce empty output; after close $fh add a check
of the child's exit status (e.g., inspect $? and die with a clear error
including $tmpfile, the command and exit code) so that when the psql invocation
(the open using "$PG_BIN/psql" with $tmpfile) fails the test fails fast instead
of using empty output for schema matching.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@sql/spock--5.0.8--6.0.0.sql`:
- Around line 1-4: The script header and user message still target 5.0.8->6.0.0;
change both to reflect the correct upgrade source 5.0.7->6.0.0 by renaming the
file comment string "/* spock--5.0.8--6.0.0.sql */" to "/*
spock--5.0.7--6.0.0.sql */" (and if the actual filename is the old form, rename
the file accordingly) and update the psql echo text that contains "ALTER
EXTENSION spock UPDATE TO '6.0.0'" to clearly reference the 5.0.7 source (e.g.,
mirror the filename change in the message so the script and its user-facing
guidance consistently indicate 5.0.7 -> 6.0.0).

In `@tests/tap/t/018_upgrade_schema_match.pl`:
- Around line 286-330: The test's catalog queries miss index definitions and
sequence properties; update the diff queries by adding a new Q_INDEXES (e.g.,
query pg_catalog.pg_indexes or join pg_class/pg_index/pg_attribute) that returns
a deterministic string per index including schemaname.table, indexname, indexdef
(or columns/order, uniqueness, predicate, expression) ordered by index name, and
extend Q_SEQUENCES to include sequence attributes (data_type/udt_name,
start_value, minimum_value, maximum_value, increment_by, is_cycled,
last_value/owned_by if relevant) so the test will detect changes like
spock.resolutions(log_time) index removal or differences in
spock.sub_id_generator AS integer/CYCLE/start; apply the same sequence expansion
where the other query copy exists (lines noted in the comment).

---

Outside diff comments:
In `@sql/spock--5.0.8--6.0.0.sql`:
- Around line 289-323: The SQL uses the regclass variable rel with format() and
%I, which quotes the whole schema-qualified name; update the format calls that
build sqlstring and ALTER TABLE to use %s for rel instead of %I: change the
first %I to %s in both SECURITY LABEL FOR spock ON COLUMN format() calls (the
ones that build sqlstring) and change %I to %s in the two ALTER TABLE format()
calls so the regclass value renders as an unquoted schema-qualified name; keep
%I for att_name and other true identifiers.