feat(orchestrator): support pi as a coding agent backend

[Bsgg1]

What

Adds pi (https://pi.dev) as a coding-agent backend for the orchestrator, alongside the existing claude/codex backends. The orchestrator invokes pi --mode json -p and reuses the operator's local ~/.pi provider settings; MetaInfer stores no API keys.

Why

next-app only supports a hard-coded claude agent. This brings the configurable agent-backend abstraction (claude/codex/pi) and the pi implementation so operators can run MetaInfer with pi as the sub-agent driver.

Changes

orchestrator/subagent_manager.py

• AgentBackend Literal gains "pi"; _normalize_agent_backend accepts pi / pi-coding-agent / earendil.
• pi_bin constructor param; _build_pi_command maps the orchestrator effort knob onto pi --thinking (low/medium/high/max) and reuses the operator's configured provider/model.
• pi session isolation: when state_dir is provided, --session-dir is scoped to <state_dir>/pi-sessions so orchestrator sessions are invisible to a user running pi --continue from their shell (pi indexes sessions by cwd; without this a user could attach to a production orchestrator session mid-flight and corrupt the context that later --session <id> --continue turns rely on). The session id is captured from pi's leading events for resume.

orchestrator/_bootstrap.py

• make_subagent_manager gains pi_bin and state_dir; both resolve from args or env (METAINFER_PI_BIN, METAINFER_AGENT_BACKEND), claude remains the default. state_dir is forwarded only to the pi backend.

orchestrator/token_budget.py

• usage_from_result_event tolerates pi's camelCase usage names (input / output / cacheRead / cacheWrite / cost.total) so a raw pi event is accounted correctly.

Tasks

• Every task orchestrator that builds a SubAgentManager now passes state_dir=state_dir so the pi backend can scope its session store per task: calc_value, find_low_hanging_kernel, gen_cpp_infer_framework, gen_infer_framework, gen_infer_framework_cpp, opt_kernel, dcu_kernel_auto_opt, evolve_kernel. opt_GEMM_kernel and port_model already wired it. Other backends ignore the kwarg.

Docs

• README (en/zh): pi backend install, env vars, and the effort→--thinking mapping.

Commits

1. feat(orchestrator): support configurable coding agent backends — cherry-picked from the existing scaffolding (author preserved), brings the claude/codex abstraction next-app currently lacks.
2. feat(orchestrator): support pi as a coding agent backend — the pi implementation on top.

Tests

• test_subagent_manager (pi command building, thinking-level mapping, session-dir scoping, resume-id capture) — 31 passed.
• test_token_budget (pi camelCase usage parsing) — pass.
• Full orchestrator + server suite: 135 passed. 2 failures are pre-existing on next-app (opt-kernel ok-detail.js missing in static/), unrelated to this PR — confirmed by stashing these changes and re-running on clean next-app.
• Task suites for touched tasks: 414 passed; 2 pre-existing failures (StateStore.init_or_resume(task_type=) API mismatch in gen_infer_framework_cpp), also unrelated — confirmed pre-existing.

No new regressions.

Usage

npm i -g @earendil-works/pi-coding-agent
pi auth status   # verify a provider is configured
METAINFER_AGENT_BACKEND=pi ./serve.py
# or if pi is not on PATH:
METAINFER_AGENT_BACKEND=pi METAINFER_PI_BIN=/path/to/pi ./serve.py