Replace the legacy toolchain with a native MkDocs lifecycle

[Alice39s]

Summary

• replace the legacy Python 3.7/pip scripts with a Python 3.14, uv-managed package and a typed Typer CLI
• replace the cache/ -> generated/ -> site/ staging pipeline with one native MkDocs plugin lifecycle reading directly from docs/
• make development rebuilds incremental through MkDocs dirty reload, cached document scanning, and changed-image processing
• generate navigation, directory indexes, contribution notices, redirects, llms.txt, Markdown copies, and optimized images without source-tree staging
• centralize assets, add WebP-first rendering and minified production HTML, and remove generator fingerprinting
• add per-page Markdown/copy actions and AI provider integrations, restore account documentation, and improve responsive article actions
• add cross-platform launchers, an 82-test suite, Ruff/mdformat checks, and Ubuntu/macOS/Windows CI

Lifecycle

Command	Behavior
uv run nmteam dev	Runs MkDocs native dirty reload directly against docs/; only changed Markdown and images are reprocessed
uv run nmteam build	Performs one strict production build directly into site/
uv run nmteam preview	Quietly previews the production output and serves Markdown as text/plain; charset=utf-8
uv run nmteam check	Runs Ruff, formatting verification, all tests, and a strict production build

The old generated mkdocs.yml, cache/, and generated/ layers no longer exist. mkdocs.yml is now the single tracked configuration source, with dynamic state supplied by nmteam-support plugin hooks.

Benchmarks

Measured on the same macOS host with isolated worktrees, identical repository content, warm dependencies, and interleaved old/new runs. Build numbers use seven runs; development numbers use three real Markdown edits; preview numbers use six alternating-order runs of 200 requests.

Scenario	Before	After	Result
Production build, median	9.41 s	4.83 s	48.7% less time / 1.95x faster
Development startup, median	9.352 s	4.856 s	48.1% less time / 1.93x faster
Development incremental rebuild, median	10.078 s	0.776 s	92.3% less time / 12.98x faster
Development output	157 lines / 26,779 B	2 lines / 254 B	98.7% fewer lines / 99.1% fewer bytes
Preview output	203 lines / 13,953 B	2 lines / 254 B	about 99% fewer lines
Preview throughput, 200 requests	1.417 s	1.419 s	effectively unchanged
Intermediate output	9,576 KiB / 440 files	none	100% eliminated

Four-way parallel image processing raises median peak build RSS from about 153.8 MiB to 235.4 MiB (+53.1%) in exchange for the build-time improvement. Two and eight worker configurations were also measured; four workers provided the preferred speed/memory balance.

Validation

• uv run nmteam check
• Ruff and formatting checks pass
• 82 tests pass on Python 3.14.4
• strict production build passes
• generated site retains 376 files and the same 15,612 KiB final footprint
• generated llms.txt matches the previous output exactly
• Ubuntu, macOS, and Windows GitHub Actions pass
• Cloudflare Pages deployment verified before opening this pull request

Merge policy

This pull request intentionally remains unmerged pending explicit approval.

[Alice39s]

Self-review verdict: changes are required before merge.

I independently replayed all 11 existing review threads and expanded the same transformation classes across the exact main@b37e5a0 → dev@84e2105 diff. The seven attached findings are new and intentionally avoid duplicating the existing footnote, bold-list, mobile-menu, and editorial-star threads.

Evidence:

• uv run nmteam check: 82 tests passed and the strict production build succeeded.
• Source-to-rendered-DOM replay found 49 incompatible hard-break rewrites across 23 pages and a list item promoted to <h2>.
• The built-site scan found three broken same-page links; one was already reported, while two more occur on the panel-launch page.
• Every local page/static target exists, all 103 emitted raster files decode, and no additional missing-asset regression was found.

The green checks currently prove buildability, but they do not catch these rendered-content regressions or silent source corruption. Please add built-output semantic coverage while fixing the findings below.

[cloudflare-workers-and-pages]

Deploying nmteam-support with    Cloudflare Pages

Latest commit:	1c43f50
Status:	✅  Deploy successful!
Preview URL:	https://d8b69e20.nmbot-telegram-doc.pages.dev
Branch Preview URL:	https://dev.nmbot-telegram-doc.pages.dev

View logs