Track progress of CREATE INDEX via pg_stat_progress_create_index.

[andreheringer]

Summary

Report diskann CREATE INDEX progress via pg_stat_progress_create_index for both sequential and parallel builds, including per-worker live tuple counts.

Changes

• Sequential build: tuples_total is set to the heap relation's reltuples, and tuples_done is updated every 1000 tuples inside build_callback_internal. The existing subphase transitions ("training quantizer" -> "building graph" -> "finalizing graph") are preserved.
• Parallel build: the blocking WaitForParallelWorkersToFinish on the leader is replaced with an interruptible observer loop. Workers periodically flush local_ntuples into a per-worker slot in DSM (allocated via a new ParallelSharedLayout wrapper). The leader sums the slots and updates tuples_done every 500ms via WaitLatch, exiting when all workers signal workers_completed.
• Progress parameter slots changed to the PG16 standard positions (subphase = 10, total = 11, done = 12).
• Python regression tests added in tests/test_progress_reporting.py (test_progress_increases, test_progress_reaches_completion).

Testing:

• cargo pgrx test (pg16): 82 passed, 10 ignored, 0 failed
• pytest tests/: 10 passed, 0 failed
• Manual live-progress probe against a freshly built pgvectorscale-debug image confirmed tuples_done rising 1024 -> 27060 during the build of a 30k-row SbqCompression index with 4 workers.

Fixes #122

[CLAassistant]

All committers have signed the CLA.