docs: add complete Agent and HTTP API tutorials

[Teingi]

Which issue or RFC does this PR close?

No linked issue or RFC.

Rationale for this change

PowerContext users do not all enter through an Agent Host. Some use Codex, Claude Code, DSH, OpenCode, or another
supported Agent. Others already have an AI application and need one language-neutral, end-to-end HTTP learning path
that goes beyond Memory CRUD.

This change makes both entry points explicit. The direct-API path now follows one coherent AI engineering-assistant
scenario across evidence capture, request-time context, work continuity, Experience and Skill evolution, Review, and
operational reporting. It also keeps identity, scope authorization, write policy, and execution authority outside the
model.

What changes are included in this PR?

• Replace the narrower bilingual Memory API quickstarts with one English and one Chinese complete HTTP API tutorial.
• Cover all 53 public OpenAPI operations across health/capabilities, Source, Memory, PreparedContext, Work, Handoff,
  Experience, managed Skill, Candidate Review, External Skill Registry, Stats, and Handoff Report.
• Provide copyable curl and jq flows for exact Source, Artifact, Memory citation, Candidate version, and Handoff
  revision handling, including optimistic-concurrency and idempotency boundaries.
• Add provider-neutral Python and model tool-calling examples that prepare read-only, untrusted context before a model
  request without exposing Server credentials or caller-selected scopes to the model.
• Show the complete Memory-to-Experience-to-Skill evolution path: direct writes, model-backed generation, pending
  Candidates, human Review, approved immutable revisions, recall, usage-based replacement, and explicit publication.
• Document the complete External Skill and Handoff Report management flows, including exact fingerprints, import/fork,
  Projects, Workstreams, Activities, retention, and Workspace bindings.
• Retain and expand the bilingual Agent step-by-step and Codex tutorials, including real Host capability differences
  and a non-Codex DSH-to-OpenCode exact-Handoff continuation.
• Align the bilingual website home pages, documentation overviews, navigation, HTTP API reference, Agent tutorial, and
  READMEs with the Agent and complete direct-API learning paths.

Are there any user-facing changes?

Yes. Users who already have an AI application can now open a single complete HTTP tutorial from the website or README
and follow PowerContext from installation through production authorization and operations. Agent users retain a
separate step-by-step entry that maps each Host to its actual Memory and Handoff surface.

The tutorial explicitly states that scope_id is partitioning rather than authorization, PreparedContext is untrusted
historical data, Candidate approval is a human-controlled CAS transition, and approved Skill content grants no tool,
file, network, secret, execution, or publication authority. There are no HTTP contract, persisted-format, package, or
runtime behavior changes.

How was this change tested?

• make docs-test — strict Zensical build completed with No issues found for the final bilingual site.
• .venv/bin/prek run --files <15 changed paths> — all applicable repository hooks passed, including TOML,
  whitespace, merge/case checks, and ty check.
• .venv/bin/pytest tests/test_api_contract.py -q -p no:cacheprovider — 26 passed, validating the canonical OpenAPI
  paths and transport models used by the tutorial.
• An automated path audit found all 53 checked-in public OpenAPI paths in both tutorials; no old tutorial links remain.
• Extracted English and Chinese bash code fences both passed bash -n; git diff --check and
  git diff --cached --check passed.
• Server-lifespan E2E tests were attempted locally but the checkout's Python 3.14 environment timed out in the first
  affected test. The available system Python 3.11 environment lacked rfc8785 and OpenTelemetry dependencies, so that
  fallback was not counted as a pass. GitHub's Python 3.11-3.14 matrix is running on the pushed head.
• Refreshed origin before push and verified local, upstream, and PR head identity after publication.

AI usage statement

OpenAI Codex (GPT-5) was used for repository inspection, documentation drafting, contract comparison, and validation.
The final commands, request/response shapes, trust boundaries, Agent capabilities, and workflows were checked against
the current OpenAPI source, Runtime implementation, integration documentation, tests, and strict site build.