feat(latency): integrate WCTT bounds into end-to-end latency (Track D commit 6/6)

[avrabe]

Summary

Closes Track D Phase 1 (FIFO + Priority networks). The latency analysis pass now alternates RTA-derived WCET on compute hops with WCTT-derived bounds on network hops, realising the WCET+WCTT end-to-end framing recorded in the project memory.

• spar-analysis::wctt::compute_network_hop_latency: new public helper returns Option<NetworkHopLatency> with (min_ps, max_ps, unservable) for a connection segment that binds to one or more Spar_Network::Switch_Type-annotated buses. Re-uses the same NC primitives (residual service, delay bound, output bound) as the existing WcttAnalysis::analyze walk.
• spar-analysis::latency: walks every connection segment in each end-to-end flow; when the helper returns Some(...), the per-hop contribution comes from the NC-derived bound and the legacy sampling-delay / inter-processor-overhead path is suppressed for that segment to avoid double-counting. When the helper returns None, the v0.7.0 path runs unchanged — that is the non-regression contract that gates Phase 1 close-out.
• Diagnostics: [network hop] annotation on per-hop diagnostics when the bound is WCTT-derived; [compute hop] annotation for same-CPU compute segments on chains that also see a network hop; LatencyHopMixed Info marker on chains that exercise both hop types.
• Unservable propagation: when compute_network_hop_latency flags a saturated residual service, latency.rs emits an Error and aborts aggregation for that chain — mirrors wctt.rs's existing WcttUnservable semantics.

What's NOT in this PR

• Phase 2 TSN-shaped service curves (TAS, CBS, frame preemption) — v0.8.x scope. Phase 2 will surface those via richer service curves in spar-network::curves and a wider Spar_TSN::* property set.
• MBPTA / stochastic NC — v0.9.0+.
• Track D Lean theorems for NC primitives (commit 5/6) — sibling PR; this commit moves forward independently because the NC Rust API is stable.
• Track E surfaces (spar moves verify deterministic-apply, MCP read-only) — Track E is parallel.

How to verify the WCET+WCTT framing on a sample model

Build a minimal cross-CPU chain with a switched bus:

bus eth
  properties
    Spar_Network::Switch_Type        => FIFO;
    Spar_Network::Output_Rate        => 1000000000 bitsps;
    Spar_Network::Forwarding_Latency => 0 us .. 0 us;
end eth;

with a system whose Actual_Connection_Binding => (reference (sw)) covers a sensor (cpu1) -> c1 -> controller (cpu2) flow. Run spar analyze and look for:

end-to-end flow 'e2e_cross' latency: [3.012 ms .. 3.012 ms]
end-to-end flow 'e2e_cross': connection 'c1' [network hop] [0.000 ms .. 0.012 ms]
LatencyHopMixed: end-to-end flow 'e2e_cross' alternates RTA-derived WCET on ...

The 0.012 ms hop is the canonical Le Boudec bound D = T + sigma/R = 0 + 1500*8*10^12 / 1 Gbps = 12 us. Removing the Spar_Network::* annotations falls back to the v0.7.0 scalar path; removing the Actual_Connection_Binding entirely also falls back. Both fall-through cases are covered by unit tests.

Test plan

• cargo build --workspace — clean
• cargo test -p spar-analysis --lib — 834 passing (824 baseline + 10 new latency tests)
• cargo test --workspace — green, no failures
• cargo clippy --workspace --all-targets -- -D warnings — clean
• cargo fmt --all -- --check — clean
• rivet validate — PASS (existing pre-existing schema warnings only)

10 new latency::tests covering: non-regression (no_spar_network_models_unchanged_v07), single-CPU chain (compute_only_chain_unchanged), single-hop NC bound matching helper (single_network_hop_uses_wctt), three-stage WCET+WCTT alternation (compute_then_network_then_compute), scalar fallback when no Switch_Type (network_hop_falls_back_to_bus_latency_when_no_switch_type), NC switchover when Switch_Type added (network_hop_uses_wctt_when_switch_type_present), LatencyHopMixed marker (latency_hop_mixed_diagnostic_emitted), unservable propagation (wctt_unservable_propagates_to_latency), unannotated-bus fallback (unannotated_bus_in_chain_falls_back), simple-fixture regression (existing_latency_fixtures_unchanged).

Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 98.41270% with 11 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/spar-analysis/src/wctt.rs	91.86%	10 Missing ⚠️
crates/spar-analysis/src/latency.rs	99.82%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!