Separate IRM default and return-type tests

[leninworld]

This PR addresses part of issue #298.

Description

Move the legacy DoubleMLIRM default and return-type tests into doubleml/irm/tests/, using the existing shared assertion helpers. Preserve the explicit IRM defaults, learner-parameter check, fitted property and nuisance-output shapes, sensitivity checks, and both policy-tree contracts. Remove the corresponding IRM coverage from the two root test modules.

The fitted IRM fixture uses the supported PSProcessorConfig(clipping_threshold=0.1) configuration. Production code and public APIs are unchanged. Other model families and the external testing guide remain outside this partial migration.

Reference to Issues or PRs

Related to #298. This PR does not complete or close the broader issue.

Validation

• Full repository suite with frozen development and RDD dependencies: 11,500 passed, 66 skipped, 3,396 warnings in 1,793.07 seconds.
• Environment: Python 3.12.13, Linux ARM64, rdrobust==2.0.0, dependencies frozen from the repository lockfile.
• Command: python -m pytest . -p no:cacheprovider --tb=short --junitxml=/results/full-suite-a2.xml.
• Skips: 48 non-single-repetition cases, 16 single-treatment-level cases, and 2 multi-instrument cases unsupported by the IV-type score.
• Focused migrated tests on commit d8248099c0fefd76eb5f34daf085701eab6314ca: 6 passed in 2.53 seconds; Ruff passed for all four changed files with --no-cache.
• Black leaves three changed files unchanged and reports only a pre-existing tuple-format difference in the root return-type module, also present at the base.
• This local result does not establish other platforms or Python versions; upstream CI results remain pending publication.

PR Checklist

• The title of the pull request summarizes the changes made.
• The PR contains a detailed description of all changes and additions.
• References to related issues or PRs are added.
• The code passes all (unit) tests.
• Enhancements or new feature are equipped with unit tests.
• The changes adhere to the PEP8 standards (Ruff passed; existing Black drift noted above).

Local validation evidence