feat(bootstrap): host performance model and cycle estimator (Wave 42, R-HS-4)

[gHashTag]

Wave 42 — R-HS-4: Host performance model and cycle estimator

Summary

Add a host-side PerformanceModel that, given engine configuration (layers, neurons, chunks), estimates:

• Per-layer inference cycles (compute + DMA)
• Total inference cycles
• Weight BRAM utilization (bytes / occupancy %)
• Estimated throughput at a given clock frequency

Scope

• bootstrap/src/host/perf.rs — new file
• bootstrap/src/host/mod.rs — add pub mod perf + re-exports
• bootstrap/src/main.rs — new CLI subcommand host-perf
• bootstrap/tests/host_perf.rs — >= 20 integration tests
• docs/NOW.md — wave-42 entry

Constraints

• Bootstrap-only (no RTL, no specs, no gen/ touched) — L2/L6 clean
• Pure arithmetic model, no hardware dependency
• All tests must pass on default cargo test

Acceptance

• PerformanceModel struct with estimation methods
• host-perf CLI subcommand
• >= 20 integration tests
• docs/NOW.md updated
• Zero regressions on existing tests