Add draft wiki-style plan for PyMC and inversion pipeline refactors

[Copilot]

This adds a concise planning document for the ongoing PyMC modelling and inversion-pipeline changes, with emphasis on the staged PyMC refactor, basis/output refactor, and how upcoming inner-domain and multi-sector work should fit into that architecture. It is intended as draft wiki content that makes the current state, migration path, and remaining work easy to track.

• New planning document

  • Adds PYMC_AND_INVERSION_REFACTOR_PLAN.md at the repository root as draft wiki-ready markdown.
  • Summarises the current status of the main refactor tracks and their relationship to adjacent work.

• PyMC refactor summary

  • Captures the Stage A-E sequence for the #370 workstream.
  • Describes the architectural shift toward:
    • canonical make_inv_inputs(...) datasets,
    • reusable builders/components in openghg_inversions.models,
    • InferenceData as the modern posterior object,
    • explicit legacy compatibility boundaries.

• Basis/output and pipeline context

  • Summarises the #360 workstream around BasisFunctions, DataTree support, legacy-output reproduction, and fixedbasisMCMC output cleanup.
  • Links that work to the high-resolution inner-region discussion and the recommendation not to extend the legacy path further for that feature.

• Migration and planning guidance

  • Adds a compact legacy-to-replacement map for the main temporary and compatibility paths.
  • Lists the highest-priority open loops, including:
    • a parallel modern inversion pathway,
    • modern post-processing inputs,
    • operator-backed basis handling,
    • CO2 multi-sector integration,
    • inner-domain alignment with the new abstractions.

• Cross-reference coverage

  • Pulls together the key umbrella issues, stub issues, the inner-region wiki page, and 2026 PRs by @brendan-m-murphy into one scan-friendly reference page.

Example excerpt from the new document:

| Legacy / temporary concept | Current or intended replacement | Notes |
| --- | --- | --- |
| NumPy-array-first `inferpymc(...)` inputs | dataset-first `inv_inputs` from `make_inv_inputs(...)` | Stage D makes the dataset-first path the main runtime. |
| Legacy trace renaming inside model construction | explicit compatibility renaming in `_rename_trace_for_legacy_inferpymc(...)` | This is the main Stage E clean-up. |