Update MAP expected values to match map_at_k in notebooks

[miguelgfierro]

Description

The notebooks were migrated from map to map_at_k and re-executed, but the tests were not updated. So the tests compare a map_at_k result against an old map expectation, and fail.

map_at_k divides by min(k, n_relevant) instead of n_relevant, so it is always greater than or equal to map. That is why every value below goes up.

This PR only moves the test expectations to the values the notebooks already produce.

Test	Notebook	Size	Old	New	MAP@K in the notebook
test_sar_single_node_smoke	sar_movielens.ipynb	100k	0.110591	0.244722	0.244722
test_sar_single_node_functional	sar_movielens.ipynb	1m	0.060579	0.185099	notebook is committed at 100k
test_sar_single_node_functional	sar_movielens.ipynb	10m	0.098745	0.203722	notebook is committed at 100k
test_baseline_deep_dive_smoke	baseline_deep_dive.ipynb	100k	0.052850	0.138672	0.138664
test_baseline_deep_dive_functional	baseline_deep_dive.ipynb	1m	0.033914	0.139016	notebook is committed at 100k
test_cornac_bivae_smoke	cornac_bivae_deep_dive.ipynb	100k	0.146552	0.330255	0.330255

Each value comes from running the notebook with the parameters the test uses. The 1m and 10m rows have no notebook value to cite because those notebooks are committed at 100k.

Only MAP moved. Every other metric in these tests still holds and is left untouched.

BPR is fixed separately in #2371, because it changed for a different reason.

Benefit

Unblocks group_cpu_002 and group_cpu_003 of the CPU nightly, which has had no green run since 2026-05-04. A real ranking regression will now surface instead of hiding behind an assertion that already fails.

Risk

• test_vw_deep_dive_smoke and test_vw_deep_dive_functional are skipped (VW pip package has installation incompatibilities), so they do not affect the nightly. Their MAP expectations are stale for the same reason as the rest of this PR, and so is the committed MAP of 0.012535 in vowpal_wabbit_deep_dive.ipynb, which predates the redefinition. Neither can be corrected until VW installs again: vowpalwabbit>=8.9.0,<9 has no cp310 or cp311 wheels, so the notebook cannot be re-run on a supported Python.
• The GPU nightly has the same stale MAP expectations, but most GPU tests override EPOCHS, so their notebook outputs are not comparable and correct values need GPU runs. test_cornac_bivae_smoke is the only one whose parameters match, and it is included here.

Checklist:

• I have followed the contribution guidelines and code style for this project.
• I have added tests covering my contributions.
• I have updated the documentation accordingly.
• This PR is being made to staging branch AND NOT TO main BRANCH.