[rksharma-owg]

JSON export reduced assumptions to their names. Reloading a model with element assumptions failed validation, while model-wide assumptions lost descriptions and excluded threat IDs. Resolved excluded findings also exported the assumption's exclusion set as a string.

This registers Assumption with the existing structured serializer and uses it for embedded assumptions. Four regression cases cover model-wide and element assumptions, including whether a threat remains excluded after exporting, importing and resolving a model. The README documents the corrected export shape.

Compatibility: exported assumptions now contain name, description and exclude, rather than a name string. The existing importer accepts these objects, and legacy model-wide string inputs remain supported. Consumers that expect exported assumptions to be strings will need to accept objects.

Validation for a941c70b917a0075297887a3b851fa8869894a8b, against 381dea05c58055ba218e632510dd3a327d2bf215:

• Local macOS / Python 3.14: baseline 251 passed; all four new cases failed on baseline; patched suite 255 passed. Scoped Black/Ruff, diff checks and source/wheel builds passed.
• Fork preflight: Ubuntu / Python 3.11–3.14 each passed 251 baseline tests, demonstrated the four expected regression failures, and passed 255 patched tests. Scoped formatting/lint and distribution builds passed on every runtime.
• Equivalent Python CodeQL analysis using v4 actions reported zero findings on both baseline and patch. The upstream workflow uses v1 actions; fork CI configuration is separate from this contribution.
• Secret-dependent optional CodeSee was not run in the fork. Upstream PR checks remain to run.

Newly discovered during local investigation; no existing issue or competing fix found.