Turn the README into a landing page, add a browser analyzer prototype

[alexandrefimov]

Summary

Two independent pieces of work on one branch; they can be split if you want the
README landing merged first.

README becomes a landing page (2979d5f). It was 843 lines, and its first
Quickstart command pointed at ./exported-impala-profile.txt, a file the reader
does not have — so the first thing a visitor tried failed, while the path that
works in ten seconds (query-doctor-web --public-demo) sat on line 597. That
path now leads. README drops to 196 lines, README.ru.md to 200. The Support
Boundary table, What Is / Is Not, Trino and Spark detail, direct-Impala history
depth, and supported deployment shape move to docs/support-boundary.md with a
Russian companion; the three entry paths move to docs/first-path.md with setup
and troubleshooting intact. The Container/Kubernetes section was already
duplicated in deploy/ and docs/recent-history-store.md, so it collapses to a
run command and links.

Eight tests asserted specific sentences against README.md, which is the
mechanism that grew it to 843 lines: every claim got pinned to the landing page.
They now assert against the document that carries the claim. Reviewers should
look at the tests/ and scripts/audit_spark_support_boundary.py diffs first —
that is the contract change, not the prose.

Also fixed: the stale claim that the 0.11.0 image will be published after the
release is approved (it has been live since 2026-08-10), and a package
description that led with four qualifiers instead of what the tool does.
Development Status moves Alpha → Beta; the version stays 0.11.0, since 1.0
touches seven files plus release metadata and belongs to the release checklist.

Browser analyzer prototype (857e783, 47263c8) under web/. A static page
that runs the deterministic analyzer in the visitor's browser through Pyodide,
so an exported profile can be diagnosed without leaving the machine. The most
common reason an operator cannot try Query Doctor is that the profile must not
go anywhere; here it does not, and DevTools shows it in ten seconds.

The analyzer core needed no changes — it imports no third-party package and no
networking or process stdlib, so it loads into Pyodide as built. Measured
against native CPython on synthetic Impala-shaped profiles: 1.9–2.2× slower,
linear at ~1.1 s per megabyte, identical output. Full page in headless Chromium:
boot 917 ms, 103 KiB profile in 116 ms, no external requests. Pyodide is fetched
at build time rather than vendored, keeping ~13 MB of binaries out of git while
the built site still serves every asset from its own origin.

Not deployed and not linked from the README. Hosting and wheel trimming are
recorded as open decisions in web/README.md.

Type

• Documentation
• Feature or workflow change
• Bug fix
• Safety or trust-boundary change
• CI, packaging, or release
• Tests only

Safety Checklist

• No raw production SQL, profiles, metadata, local paths, hostnames, secrets, subprocess output, model/runtime internals, or raw artifact filenames are introduced.
• Browser-visible UI and trusted reports remain sanitized.
• External collection changes, if any, are explicit, bounded, read-only, and redacted by default. (None; the new page performs no collection and no network access after load.)
• Query Optimizer changes, if any, do not execute user SQL or optimizer draft SQL. (None.)
• LLM-facing changes, if any, keep Python/analyzer facts as the source of truth. (None.)
• Unsupported evidence remains unknown, not_observed, or explicitly unsupported.
• No screenshots, logs, generated reports, or browser captures from real clusters are added.
• Fixtures, examples, and test names use synthetic placeholders only. web/bench/make_profile.py generates profiles with example.invalid hosts and synthetic_* identifiers.

No support claim changes. The moved text is verbatim; 50 of the 52 body lines in
docs/i18n/ru/support-boundary.md are byte-identical to the already-public
README.ru.md on main, the other two being relative-link fixes.

Validation

• Full suite: 4697 passed, 1 skipped
• Docs CI: check_active_docs, audit_public_docs, audit_public_distribution_boundary, check_markdown_links
• scripts/check_staged_public_safety.py --changed over the whole branch diff: OK
• python3 scripts/agent_preflight.py
• ruff check query_doctor scripts tests web/bench
• git diff --check
• scripts/installed_readme_quickstart_smoke.py against a clean wheel install, exercising the three commands the README now shows
• query-doctor-demo-preflight --public-release: READY_WITH_WARNINGS
• pre-commit run --all-files — not run locally; CI covers it
• Release gate — not release-facing

The preflight gained four warnings on docs/first-path.md for naming
analysis.json, analysis_facts.md, profile.txt, and CM_PASSWORD. That is
moved text, the scan only inspects changed files, and docs/local-smoke.md and
docs/credentials.md already carry the same tokens on main.

Documentation

• Public docs are updated.
• Documentation drift was checked for changed behavior, contracts, commands, routes, and safety wording.
• docs/changelog.md is updated.
• New examples and fixtures are synthetic and sanitized.

Branch Hygiene

• This PR targets a review branch, not direct pushes to main.
• Branch is based on current main.

---

Assisted-by: Claude Opus 5 via Claude Code