fmakv2 version added to fma_keys

[campeasb]

fma_keys: add support for v2 metadata (backward-compatible)

One-liner:
Extend the existing fma_keys dataset to support the v2 metadata/index while keeping full backward compatibility with v1.

---

Description

This PR extends the existing fma_keys loader to handle a v2 metadata format and index.
There is no new dataset module; changes are limited to fma_keys internals + its top-level docstring.

---

What’s included

• Support for the v2 CSV/metadata schema (track id, key, mode, optional Spotify URI, etc.).
• Backward-compatible handling of both v1 and v2 indices/metadata (auto-detection based on schema).
• Internal helper(s) for parsing/normalizing v2 fields (keys/modes mapping remains consistent across versions).
• Docstring update in fma_keys to clearly document:
  • Annotations: track-level key & mode (e.g., C…B; major/minor).
  • Dataset size: ~5,489 tracks (~340 hours), 17 genres.
  • Origin: Free Music Archive subset annotated by an expert musician (perfect pitch, ~20y experience); small cross-annotation subset with high agreement.
  • Music types: multi-genre FMA content.
  • References: include FMA dataset reference (Defferrard et al., 2017) and any dataset notes relevant to key annotations.
  • Access & license: available via Zenodo; non-commercial research use.

Note: Figures above reflect the established FMA Keys dataset characteristics; docstring now explicitly mentions v1 and v2 formats and how they’re handled.

---

Changes (code)

• Extend mirdata/datasets/fma_keys.py to parse v2 metadata/index.
• Preserve existing public API and outputs (no breaking changes).
• Small refactors for clarity around metadata normalization (keys/modes).

---

Tests & coverage

• No new test files added.
• Existing fma_keys tests continue to pass locally with v1.
• Local runs confirm v2 parsing paths (where applicable) without impacting coverage (CI will validate).

---

Documentation

• Top-level docstring of fma_keys updated to:
  • Describe both v1 and v2 metadata formats.
  • Include size/origin/usage/license info.
  • Link to canonical access (Zenodo) and relevant references.

---

Dataset loader checklist (adapted)

• ✅ New script under scripts/… (no new dataset module added in this PR).
• ✅ New index upload fmakv2_index.1.0.json
• ✅ Sample/index handling remains aligned with fma_keys.
• ✅ Loader lives in mirdata/datasets/fma_keys.py (extended, not duplicated).
• ✅ Tests: no new files; existing tests pass (CI to confirm).
• ✅ Docs: updated module docstring; table entries unchanged (no new dataset).
• ✅ Formatting & typing: black, flake8, mypy (CI to validate).
• ✅ tests/test_full_dataset.py on fma_keys (local sanity; CI to validate).
• ✅ Codecov: no expected decrease.

---

Notes

• The PR focuses strictly on versioning support within fma_keys (v1 + v2), not on introducing a separate dataset.
• Public behavior is unchanged; v2 is parsed transparently.
• If maintainers prefer, feel free to make small commits on top of this PR to streamline the merge (remove with please-do-not-edit if undesired).