Preserve sample_query_options during cohort query normalisation

[kshirajahere]

Summary

_setup_cohort_queries() still drops sample_query_options when it re-applies cohort queries during cohort-size validation, so parameterised sample filters like country in @countries_list fail in any API that relies on shared cohort normalisation.

Why this matters

This is the same correctness family as #995 / #999, but at a deeper layer: the recent caller-level forwarding fixes in #996 and #1001 do not cover the helper that actually rebuilds and rechecks cohort queries.

As a result, higher-level analyses that accept sample_query_options still break for valid pandas query context such as:

• local_dict
• global_dict
• resolvers

That matters for multi-cohort analysis surfaces because it makes query behaviour inconsistent depending on which code path is used, even when the public signature advertises sample_query_options support.

Reproduction

Using the simulated-data APIs locally, both of these currently raise pandas.errors.UndefinedVariableError: local variable 'countries_list' is not defined:

countries_list = ["Angola", "Mali"]

api.pairwise_average_fst(
    region="2L",
    cohorts="country",
    sample_query="country in @countries_list",
    sample_query_options={"local_dict": {"countries_list": countries_list}},
    min_cohort_size=1,
)

api.plot_h12_gwss_multi_overlay(
    contig="2L",
    cohorts="country",
    window_size=200,
    sample_query="country in @countries_list",
    sample_query_options={"local_dict": {"countries_list": countries_list}},
    min_cohort_size=1,
    show=False,
)

The failure happens inside malariagen_data/anoph/sample_metadata.py when _setup_cohort_queries() loops over derived cohort queries and calls:

self.sample_metadata(sample_sets=sample_sets, sample_query=cohort_query)

without passing sample_query_options.

Additional mismatch in the same query contract

Once sample_query_options is forwarded into the shared helper, sample_metadata() also needs to treat engine="python" as a default rather than forcing it as a duplicate keyword argument. The current implementation documents engine as a supported query option, but paths that now correctly forward sample_query_options={"engine": "python"} can trip:

TypeError: DataFrame.query() got multiple values for keyword argument 'engine'

Proposed fix direction

1. Pass sample_query_options through the cohort-size validation calls inside _setup_cohort_queries().
2. Normalize query options in one shared place so engine="python" remains the default, but an explicitly supplied engine does not become a duplicate keyword.
3. Add regression tests at both the metadata layer and at least one multi-cohort public API surface (e.g. H12/Fst) using local_dict.

Impacted files

• malariagen_data/anoph/sample_metadata.py
• malariagen_data/anoph/base.py

This looks like a good small-but-deep correctness fix because one shared abstraction affects multiple analysis entry points.