Use current PyEnsembl and gtfparse with explicit pandas 3 coverage

[iskandr]

Require PyEnsembl 2.24.1+ and gtfparse 2.9.1+, and explicitly use gtfparse 3.0.2+ with pandas 3 in CI. Support pandas 2.2.2+ and 3.x while retaining Python 3.9/3.10 and joint Topiary installations. Topiary's current gtfparse <3 cap is tracked separately in openvax/topiary#484; the compatible 2.9.1 floor avoids creating a resolver conflict.

CI selects dependency lines explicitly: Python 3.9/3.10 with pandas 2 and gtfparse 3; Python 3.11 with pandas 2/gtfparse 2.9.1 and pandas 3/gtfparse 3; Python 3.12 with pandas 3/gtfparse 3. Each job validates the installed source distribution and dependency consistency; cross-annotator parity runs with pandas 3. Document the upgrade command and bump Varcode to 10.11.8.

Validation: ./lint.sh, git diff --check and pip check pass. On the final branch, ./test.sh passed all 3,240 tests (92% coverage), and ./test.sh --annotator=protein_diff -m parity passed all 3 parity tests, using Python 3.12.6, PyEnsembl 2.24.1, gtfparse 3.0.2, pandas 3.0.6 and Sercol 1.0.3. The five-job GitHub CI matrix and docs build gate merging.