feat: add qualified Popcount and round-robin PYC primitives

[HKU-LiBoyu]

Summary

This PR adds the pyCircuit implementation half of the qualified primitive vertical slice paired with PTO-ISA/agentic-circuit#18:

• Add pyc.popcount and pyc.rr_arbiter to the PYC MLIR dialect with verifier checks for integer widths, legal popcount result width, arbiter request/grant width, cursor width, and the supported 2..64 input range.
• Extend the C++ emitter with pyc::cpp::popcount<OutWidth, InWidth> and pyc::cpp::rr_arbiter<NumInputs, PointerWidth> calls.
• Extend the Verilog emitter and pycc primitive packaging so generated RTL instantiates the parameterized primitives and includes their source files.
• Add reusable C++ runtime helpers and synthesizable Verilog implementations for population count and non-power-of-two-safe round-robin one-hot arbitration.

Files changed

• PYC dialect definition/verifiers: compiler/mlir/include/pyc/Dialect/PYC/PYCOps.td, compiler/mlir/lib/Dialect/PYC/PYCOps.cpp
• C++/Verilog emitters: compiler/mlir/lib/Emit/CppEmitter.cpp, compiler/mlir/lib/Emit/VerilogEmitter.cpp
• Primitive packaging: compiler/mlir/tools/pycc.cpp
• C++ runtime: runtime/cpp/pyc_primitives.hpp
• Verilog RTL: runtime/verilog/pyc_popcount.v, runtime/verilog/pyc_rr_arbiter.v

Contract

• pyc.popcount: integer input width W produces ceil(log2(W + 1)) bits.
• pyc.rr_arbiter: packed req[N-1:0], packed cursor[ceil(log2(N))-1:0], one-hot grant[N-1:0], with 2 <= N <= 64; invalid cursor values normalize to zero for non-power-of-two N.
• Cursor state remains owned by the surrounding PYC graph; the primitive is combinational.

Validation

• C++ smoke: N=4 (64 vectors), N=3 (32 vectors), all passed.
• Verilator exhaustive simulation and lint: N=4/N=3 passed.
• Yosys prep/check/stat: 0 problems.
• LLVM 22 mlir-tblgen PYC dialect syntax check: passed.

The paired agentic-circuit lowering and ACIR fixtures are in PTO-ISA/agentic-circuit#18.