Phase 6: 9/7 irreversible DWT + performance tuning + multi-component

- Add FusedStripJob97 for 9/7 wavelet: calls cascade_strip_97 which
  performs fused H-DWT + cascade V-DWT with partial output per strip
- Remove qmfbid!=1 exclusion: ScxEngine now handles both 5/3 and 9/7
- Use appropriate halo per wavelet type (1 for 5/3, 4 for 9/7) in
  StripPartitioner::partition calls
- Performance tuning: computeIdealStripeHeight now accounts for halo
  overhead (targets 384KB working set), and ensures enough strips for
  parallelism (minimum 2×nproc strips per resolution)
- Multi-component: document that ScxEngine naturally interleaves DWT
  strips from all components via independent per-component domains

Author: Grok Compression