Queue Theory: MMc (split & FCFS/Erlang-C), independent Observed KPIs, richer sampled metrics, and updated notebooks

[GioeleB00]

Description
This PR upgrades AsyncFlow’s queue-theory analysis and runtime metrics so users can compare Theory vs Observed in a transparent, model-faithful way across M/M/1 and M/M/c scenarios.

Highlights

• MMc support (two flavors)

  • Random split (M/M/c → c parallel M/M/1): closed forms with per-server arrival λᵢ = λ/c.
  • Pooled FCFS (Erlang-C): classic formulas (P₀, Pᵥ, Wq, etc.) surfaced as the “FCFS/Erlang-C” mode.

• Observed KPIs computed independently

  • λ̂ from throughput series (mean RPS), μ̂ from per-request service_time.
  • Ŵ from client latencies; Wq̂ from LB waiting times (FCFS) or server waiting_time (split).
  • L̂ from L_SYSTEM series (fallback to λ̂·Ŵ only if missing).
  • Lq̂ from LQ_LB (FCFS) or the sum of per-server LQ_SERVER means (split); fallback to λ̂·Wq̂ only if missing.
  • ρ̂ from SERVER_UTILIZATION series (fallback to λ̂/(c·μ̂) only if needed).
  • No circular derivations: each KPI stands on its own measurement.

• New/expanded sampled metrics (exposed by the analyzer)

  • L_SYSTEM (system population at the generator scope).
  • LQ_SERVER and LQ_IO (per-server queues).
  • SERVER_UTILIZATION (busy indicator per server).
  • Existing maps are accessible uniformly via get_metric_map(...) (enum or string key).

• ResultsAnalyzer quality-of-life

  • get_server_event_arrays() now exposes per-request arrays:
    latencies, service_time, io_time, waiting_time, finish_times.
  • list_server_ids() helper for stable server enumeration.
  • Per-server throughput series via get_server_throughput_series(...).

• Comparison tables

  • Pretty Theory vs Observed tables for MMc (split and FCFS/Erlang-C) and MM1.
  • Clear headings, deltas, and ∞ handling when ρ ≥ 1.

• Notebooks & docs

  • Updated guided notebook for MMc split with end-to-end build-run-compare flow.
  • Added FCFS/Erlang-C comparison path and a quick introspection snippet to verify that
    time-series are present and used (no silent fallbacks).