Prototype pairwise seeding and optimize find-orientations

[joelvbernier]

Summary

This PR tightens the performance and robustness of the legacy find-orientations workflow and also prototypes a new seeded indexing path based on pairwise continuous fiber intersections (work in progress).

What changed

• added pairwise and pairwise-greedy orientation candidate generators that avoid rasterizing continuous fibers into dense point clouds before scoring
• added Friedel-pair condensation ahead of candidate generation so duplicate eta-omega seed peaks are reduced before discrete fibers are generated
• made Friedel condensation the default for seeded searches, while keeping it configurable
• added exact post-hoc quaternion misorientation merging for sparse greedy candidates
• added a synthetic eta-omega scaling benchmark harness based on simulateOmeEtaMaps()
• refactored multiprocessing in the legacy path to use HEXRD's platform-aware multiprocessing context
• fixed paintGrid chunk sizing for small jobs
• added spawn-aware serial fallback for fine-grained pools on macOS/Windows, where process overhead dominated runtime
• added a paintGrid eta/omega tolerance floor to the map bin width, with warnings when the requested tolerance is below map resolution

Why

The original seeded workflow was robust but wasteful: it rasterized many candidate orientations along each fiber, scored them all, and relied on clustering to recover the true modes afterward.

This PR moves toward a more economical seeded-forward-model approach while also improving the existing discrete-fiber path:

• reduce redundant seeds early via Friedel pairing
• preserve cross-platform behavior under Linux/macOS/Windows multiprocessing constraints
• keep paintGrid and downstream scoring physically consistent with the eta-omega map resolution

Benchmark notes

On the multiruby eta-omega benchmark with the legacy discrete-fibers path:

• 1 CPU, Friedel off: 14.314 s, 71,281 candidates, 3 grains
• 1 CPU, Friedel on: 10.264 s, 38,520 candidates, 3 grains
• 4 CPU, Friedel on (macOS spawn context): 10.188 s, 38,520 candidates, 3 grains

So the Friedel condensation clearly wins in the legacy path, and the spawn-aware fallback prevents multiprocessing from becoming dramatically slower on macOS.

Testing

• python -m unittest tests.test_indexer tests.config.test_find_orientations

Follow-up

• Profile the legacy path on a fork-based Linux environment to confirm the multiprocessing upside there
• Continue iterating on the pairwise-greedy path now that the legacy baseline is stronger and better characterized

[psavery]

This needs to be updated to use the simpler formulas that Don determined over the past couple of weeks.