AlphaPulldown v2.11.0

What's Changed

• --allow_resume now works for AlphaFold 3: a rerun skips finished folds and continues an interrupted one seed by seed, in single runs and in batches. Each fold keeps a record of how it was predicted, so a resume never mixes settings; AlphaFold 2 resumes now keep finished relaxations as well.
• Fold directories keep AlphaFold's outputs at the top and AlphaPulldown's own files in two subdirectories: alphapulldown/ (the fold record, the inference-kernel record and the copied feature metadata) and debug/ (what --debug_msas and --debug_templates write). Directories from 2.10 keep working.
• Backend options are named explicitly, so a mistyped option is reported instead of silently falling back to its default.
• Fix AlphaFold 2's ModelCIF conversion for output paths with spaces or shell characters, and report its real exit status instead of treating any log output as an error. AlphaFold 3 no longer writes template debug files into the working directory, and a fold left without ranked samples is no longer recorded as finished.
• The AlphaFold 3 backend is split into input, output, runner and debug modules, with prepare_input returning PreparedFold objects; predictions are unchanged.

Full Changelog: 2.10.0...2.11.0