fix(btw): record runner errors and refused work runs

[YUZHEthefool]

Summary

Work runs that errored, or never ran at all, reported completed. Expected a
local provider error, a third-party runner error, and a request refused by
session admission to be failed.

Related issue

Fixes #154

Root cause

The work loop read a single failure marker, and nothing set it on these paths:

• run_agent turns an err response into a user-facing error result and then
  drains its generator normally, which the loop cannot tell apart from an answer.
• ThirdPartyAgentSubStage records _third_party_runner_error and nothing read it.
• AgentRequestSubStage returns before building an Agent when the provider is
  disabled or the session has AI off, so the executor yielded nothing and the
  generator ending looked like success.

Reproduction

Enable BTW and the work loop, then (a) drive the real run_agent with a runner
that returns err, (b) drive the real ThirdPartyAgentSubStage non-streaming
handler with a runner that raises, and (c) drive the real AgentRequestSubStage
with session AI disabled. All three used to read completed. Each is a
regression test.

Implementation notes

Write the failure marker where the failure happens and read both markers in the
loop. The marker keys move into btw/types.py, which imports only the standard
library, so the writers (agent stages, run_agent) and their only reader cannot
drift apart; third_party.THIRD_PARTY_RUNNER_ERROR_EXTRA_KEY still resolves, now
by import. mark_work_run_failed keeps chat runs on their existing error path.

An executor that produced nothing also no longer counts as completed: a generator
ending only proves the task ran when something actually ran. That is the
"do not equate normal generator termination with success" criterion, kept narrow
enough that queued work is unaffected — it is owned by the semaphore, not by
this generator.

Validation

.venv/Scripts/python.exe -m pytest tests/unit -q
  5219 passed, 6 skipped

.venv/Scripts/python.exe -m ruff format --check <touched files>   # clean
.venv/Scripts/python.exe -m ruff check <touched files>            # clean
.venv/Scripts/python.exe -m pyright --project pyrightconfig.quality.json <touched files>
  0 errors

All five new tests were run against the unmodified tree first and failed with
WorkSessionStatus.COMPLETED, so they are real regression tests rather than
descriptions of current behavior.

make check, make quality, and the coverage-gated --test-profile all run
were not run locally on this Windows checkout; this PR relies on CI for those,
as #168 did.

After #168 was rebased onto the updated master head, this branch was rebased onto
the new head as well and the whole stack was re-validated at its tip:
pytest tests/unit -q 5233 passed / 6 skipped and
pytest --test-profile blocking -q 5354 passed / 6 skipped / 1 deselected.
The count above was taken on this branch before that rebase.

Compatibility and risk

• No route, schema, configuration, or dependency change. BTW stays off by default.
• THIRD_PARTY_RUNNER_ERROR_EXTRA_KEY keeps its name and value; its definition
  moved from third_party.py to btw/types.py.
• A work run that reaches no Agent is now failed instead of completed. That
  is the intended behavior change.

Checklist

• The change is focused and does not include unrelated refactoring.
• I added or updated a regression test, or explained why a test is not practical.
• I ran the relevant formatting, lint, build, and test commands.
• User-visible behavior updates both docs/zh/ and docs/en/ when needed. (No user-visible text changed.)
• OpenAPI, generated client, docs/public/openapi.json, and tests change together when routes or schemas change. (No route or schema change.)
• No secrets committed. Runtime Python deps update pyproject.toml, requirements.txt, and uv.lock together. (No dependency change.)
• I did not restore legacy shims, Python <3.14 fallbacks, or upstream publish/docs URLs as fork artifacts.
• Breaking API or behavior changes use ! and a BREAKING CHANGE: footer. (None; the status change is the fix.)
• I will not merge this PR myself. Merge needs a human maintainer review plus a separate AI-assisted review (AI_POLICY.md).
• AI use follows AI_POLICY.md. Keep exactly one author note below. Do not fabricate the other.

Agent note

Goal: resolve #154 with an explicit terminal outcome instead of
a bare generator end, keeping BTW off by default and preserving authorization and
request identity.

Paths touched: astrbot/core/agent/btw/types.py,
astrbot/core/agent/btw/work_loop.py, astrbot/core/astr_agent_run_util.py,
astrbot/core/pipeline/process_stage/method/agent_request.py,
.../agent_sub_stages/internal.py, .../agent_sub_stages/third_party.py, and
unit tests (test_btw_work_loop.py extended, test_agent_request_sub_stage.py
extended).

Checks run: the commands under Validation, plus the new tests against the
unmodified tree.

Residual risk: run_agent marks a work run failed on any err response, even if
a later step were to succeed. An err already produced a user-visible error, so
reporting the run as failed is the honest reading.

Tools used: Claude Code (Opus 5) with the repository's AGENTS.md and AI_POLICY.md.

This PR is stacked on #169; merge that first.