Align reference architecture with P.A.I.H.I. and add a live PAIHI Scorer

[altmanAI]

Authorized by Blake Hunter Altman — ALL CLEAR FOR IMPACT (2026-07-08).

What this does

This repo already had a real, clean, working reference implementation — orchestrator, values engine, memory store, transparency log, tests, docs. But it defined its own five-principle framework ("Consent over assumption", "Values as code, not vibes", etc.) instead of AltmanAI's actual P.A.I.H.I. framework (Proof, Alignment, Integrity, Humanity, Impact) — a brand collision between two competing "what is human-first AI" definitions under the same org.

This PR:

• Remaps the architecture explicitly onto P.A.I.H.I. (README + docs/ARCHITECTURE.md)
• Adds human_first_ai.paihi.PAIHIScorer — computes a real 0-100 score per P.A.I.H.I. dimension from a completed pipeline run, so the framework is operational code, not just a label
• Wires Orchestrator to track RunRecords (decision, checkpoint offered/honored, completed) so the scorer has real signal
• Updates demo.py to print a live P.A.I.H.I. Score at the end of the walkthrough
• Adds 5 new tests (20 total, all passing)

Verification

pytest -q
....................
20 passed

python -m human_first_ai.demo
...
--- P.A.I.H.I. Score for this session ---
P — Proof:      100.0/100
A — Alignment:  100.0/100
I — Integrity:  100.0/100
H — Humanity:   100.0/100
I — Impact:      66.7/100
----------------------------------------------------
Overall P.A.I.H.I. Score: 93.3/100

No new repo created — consolidating into this existing one to avoid further fragmentation across the org (human-first-ai / altmanai-sdk / PAIHI-Portables already overlap; flagged separately for a future consolidation pass).