Fix AV network sky-frame inverse mapping

[oshaughnessy-junior]

Summary

Fix --internal-sky-network-coordinates in batch AV by mapping sampled network-frame sky angles back to physical equatorial coordinates with the inverse frame before likelihood evaluation and XML serialization.

Previously the driver reused the forward physical-to-network frame in both places. That applies the same rotation twice rather than undoing it, so the Cartesian AV coordinates were not the intended detector-baseline coordinates.

This change also adds explicitly named physical-to-network and network-to-physical helpers to make the direction unambiguous.

Regression coverage

• Check that physical-to-network conversion gives cos(theta_HL) = n dot baseline at a fixed epoch.
• Check that network-to-physical conversion round-trips two representative sky directions.
• The batch driver now calls the tested network-to-physical helper for both likelihood evaluation and CPU/XML export.

Validation

On citlogin6 in the IGWN Python environment:

2 passed, 4 warnings in 2.58s

The warnings are the pre-existing NumPy matrix deprecation from assign_sky_frame.

Found while diagnosing S250114ax AV sky sampling; this PR changes only the coordinate mapping, not the likelihood or sky prior.

[interhost-cross-review-canary-v2]

Independent automated review completed at the recorded exact commit. Detailed findings were withheld from public output by the private-context egress policy and require private human declassification.