Dependency-gate guard cannot distinguish PyPCAP from PCAP_CT, so a wrong install is a false pass #762

Describe the bug

tests/_dependency_gates.py's liveness guard cannot tell PyPCAP from PCAP_CT, so a CI job
installing one while reaching the other's gate is reported as satisfied rather than as a gap.
Both distributions install a top-level pcap module, and extras_providing() resolves either
import name to both extras.

Reproduction

import _dependency_gates as g
print(sorted(g.extras_providing('pcap')))        # ['PCAP_CT', 'PyPCAP']
print(sorted(g.extras_providing('pcap._pcap')))  # ['PCAP_CT', 'PyPCAP']

Doctoring one install line each way, against the real guard:

It is blind in both directions. Liveness still passes in both worlds because HAS_PYPCAP's
exclusion lists integration+gate, which are real gaps — so the wrong non-gap is never examined.

Expected behavior

A job installing PCAP_CT should be reported as not satisfying HAS_PYPCAP, and vice versa.

Additional context

The conflation is not at the flag level. flag_requirements() resolves them apart correctly —
('tests/foundation/engines/test_new_engine_parity_runtime.py', 'HAS_PYPCAP') -> {'pcap'} and
('tests/foundation/engines/test_pcap_ct_engine.py', 'HAS_PCAP_CT') -> {'pcap._pcap'}. It is
extras_providing() one level down that collapses both.

Proposed fix, same shape as the existing machinery and no per-distribution model needed: assert that
every (job, gate) pair deemed satisfied is satisfied through an import name with exactly one
providing extra, else require an explicit allowlist entry. That fails loudly on both rows above.

Found while cross-reviewing #755; the guard itself landed in #745.