fix(core): resolve analyzer perf regression and harden pirx-core visibility

[m2papierz]

What

Fix ~20-25% performance regression in ProfileAnalyzer::analyze introduced by error budget tracking, apply the same fix to monte_carlo::extract_summary, and tighten module/type visibility across pirx-core.

Why

CodSpeed flagged analyze[500] at -25.49%, analyze[100] at -22.85%, analyze[10] at -19.7% comparing master
(71f67ed) against the error budget baseline (8bc4d1a). The regression was caused by commit 92cc496 restructuring the GateServed match arm from a guarded pattern to an unguarded one, making every zero-wait event pay for a
saturating_add + bounds-checked array write it previously skipped. The same redundant-counter pattern existed in monte_carlo::extract_summary. Several internal-only types and modules were unnecessarily pub, widening the API surface beyond what downstream crates consume.

How

Analyzer (analyzer.rs):

• Restored the guarded match arm (GateServed { gate, wait } if *wait > 0) for the stall+counting path; added a
  separate GateServed { .. } arm for zero-wait bucket counting only.
• Replaced the redundant total_magic_states loop counter with trace.magic_states_consumed (already computed by the engine).
• Fused the post-loop cumulative_magic_states prefix-sum and cumulative_infidelity computation into a single
  pass, eliminating a separate .map().collect() allocation.

Monte Carlo (monte_carlo.rs):

• Removed the redundant magic_states_consumed counter; reads trace.magic_states_consumed directly.
• Restored the GateServed guard so zero-wait events fall through to the no-op arm.

Visibility (lib.rs, buffer.rs, events.rs, routing.rs):

• buffer, events, routing modules: pub → pub(crate).
• ScalarRouting, ManhattanRouting, RoutingKind, build_position_index, from_config: pub → pub(crate).
• MagicStateBuffer, EngineEvent, TimedEvent, EventQueue: pub → pub(crate).
• MagicStateBuffer::is_full, is_empty, EventQueue::is_empty: gated behind #[cfg(test)] (test-only usage).

Testing

• make ci passes locally (fmt + clippy + test + audit)
• New behavior has tests (existing 206 tests cover all changed paths)
• Hot-path changes have criterion benchmarks (trace_analysis/analyze/{10,100,500})

Checklist

• PR description explains why, not just what
• No new unwrap()/expect() in production code
• No new allocations in the simulation hot loop
• Crate boundaries respected (pirx-core never imports from pirx-adapters)
• New dependencies justified (not "it's popular" — what does it replace?)

[codspeed-hq]

Merging this PR will improve performance by 19.31%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
✅ 12 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	analyze[500]	376.4 µs	314.1 µs	+19.84%
⚡	analyze[100]	85.8 µs	71.6 µs	+19.77%
⚡	analyze[10]	16.3 µs	13.8 µs	+18.32%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing fix/analyzer-perf (3711630) with master (71f67ed)}