0f626eed

By: ValentinHirschi <valentin.hirschi@gmail.com>

run sampling acceptance through ordinary integration

Expose a validated Gaussian reference descriptor on the integrate command and
Python API, using the existing process maps, native stability stack, worker
slots, adaptive grids and complex accumulators. Report normalization and the
normalized raw momentum-squared moment as two observables of expectation one.
Persist the descriptor and reporting convention in workspace manifest version2
and reject physical/reference mismatches before restoring settings. Existing
integration workspaces require --restart; generated gammaloop_state data and
the binary integration checkpoint layout are unchanged.

User prompts:
"Continue as planned, but two more points:

a) You need a systematic and generic testing harness for these new maps. On top of more targeted unit tests, I suggest to design this harness as an acceptance test too being able to take a state with its own definition of complicated maps from an actual graph (and also with potentially multi-channeling enabled), and then just substitute the final weight of the integrand to be that of a simple multidimensional normalized function over the infinite spatial momenta volume (something like a gaussian with width similar to sqrt(s) or anything else that you deem appropriate) and verify that you integrate to 1. This way you can easily systematically use this harness for a wealth of different map setups.

b) You did not tell me how you plan on addressing the case where depending on the cut momenta, an e-surface exists or does not exist. How would you be able then to properly sample the surface to the left or right of that cut, with a proper jacobian, if it does not always exist?"

"Just a status question—please continue your current orchestration unchanged: where does the joint two-normal density sit in the implementation sequence, and what prerequisites remain before you reach it? (direct-H focusing alone is not expected to bound the previously identified corner’s weights.)"

Validation:
- Core/API/Python cargo check and all-target clippy pass; new argument-count
  warnings resolved. Existing Python helper warning remains (24 to25 arguments).
- 43 focused core/API checks pass, including all three loaded-state reference
  fixtures, exact checkpoint resume and actual unequal trained-grid weights.
- 53 normal core integration-owner checks pass. The exploratory default-profile
  run also hits the pre-existing failing:: zero-relative-reference ETA test;
  its unchanged assertion fails and curated repository profiles exclude it.
- All11 differential and6 multi-integrand physical/model/batching/resume tests
  pass using nextest profile test_gammaloop. Test-target check/clippy pass.
- Two stale CutInfo.lmb_channel_id test-helper accesses updated to the existing
  canonical sampling_channel_id so that test target compiles.
- cargo fmt --all -- --check and git diff --check pass.