refactor(pyc): unify zero-count primitive family

[zhoubot]

Summary

• keep simple public count_leading_zeros and count_trailing_zeros helpers
• unify canonical ACIR and PYC into one count_zeros operation with verified static leading|trailing direction
• freeze direction in QueueGraph plans and select the corresponding gfsim helper during C++ generation
• replace the leading-only gfsim block with one CountZeros<Width, Direction> SimQueue template plus readable aliases
• replace the leading-only RTL with one repository-owned BSD-3-Clause balanced tree parameterized by DIRECTION_LOW
• emit two manifest bindings of one implementation identity; width and source selection remain fail closed

Decision

Decision 0166 records that direction is a semantic compile-time parameter, not a second implementation-specific API or runtime value.

Validation

• pre-commit run
• pytest tests/unit -m unit — 23 passed
• pytest tests/system/test_primitive_selection.py -q — 16 passed
• Agentic Python frontend/JIT/tool lane — 144 passed, 2 environment skips
• exact Agentic CI unittest lane — 37 contracts tests plus 141 frontend tests passed (3 environment skips)
• check-acir — 141/141
• ACIROpsTests — 1845/1845
• CodeGenTests — 98/98
• GfsimTests — 252/252
• repository contracts, IR coverage, API hygiene, and MkDocs strict — passed
• decision-status validation — 166 rows, 0 deferred
• independent review — APPROVE

Evidence: docs/gates/logs/20260905-pyc-zero-count-family-r1/summary.md

Stack

This PR is intentionally based on #37 and should be reviewed/merged after the earlier stacked PRs.