architecture page: explain how the engine works + add diagrams

The previous architecture page was a curated list of pointers
into the audit-docs corpus. Rewrite it as an actual deep-dive
that explains how the engine produces records, with Mermaid
flow diagrams for the load-bearing concepts.

Sections added:
- The mental model in one diagram (DSL → plan → executor → IR → record)
- How the DSL becomes a simulation (full lowering pipeline)
- The seven engine invariants in user-friendly tone with
  diagrams where they help (constrain-before-propose, replay
  admissibility chain, live-call refresh)
- Three streams per pass (trace / events / Simulation)
- Plan signature replay gate (three-stage check)
- Two-layer integrity model (AIRR validator + cache parity)
- Audit-first workflow

Enable Mermaid via pymdownx.superfences custom_fences + a
small init script that loads mermaid from CDN as ESM. The
init also re-runs on Material's instant-nav swaps so diagrams
render after client-side navigation.

Audit-docs deep links switched from relative paths to GitHub
absolute URLs (audit-docs/ is at repo root, outside the MkDocs
docs_dir, so relative paths can't resolve in the built site).

Author: MuteJester

mkdocs.yml

@@ -82,6 +82,13 @@ theme:
 extra_css:
   - stylesheets/extra.css
 
+# Mermaid for architecture flow diagrams. Loaded as an ES module from
+# CDN so we don't ship the (~1MB) library locally. The pymdownx
+# custom_fences config above turns ```mermaid blocks into
+# <pre class="mermaid"> elements; mermaid.run() picks them up.
+extra_javascript:
+  - js/mermaid-init.js
+
 markdown_extensions:
   - admonition
   - attr_list
@@ -92,7 +99,11 @@ markdown_extensions:
       permalink: true
       toc_depth: 3
   - pymdownx.details
-  - pymdownx.superfences
+  - pymdownx.superfences:
+      custom_fences:
+        - name: mermaid
+          class: mermaid
+          format: !!python/name:pymdownx.superfences.fence_code_format
   - pymdownx.snippets
   - pymdownx.tabbed:
       alternate_style: true