feat(plugins): add AuthorityRoutingPlugin — pre-execution authority posture (ADVISE/EXECUTE/DEFER/STOP)

## Problem

Tool schemas and permissions govern *access* — which tools exist and who may call them. They do not govern *authorization*: whether the agent was allowed to act **at all** for a given request, and within what scope. That gap is where three recurring agent failures live:

- **advice becomes action** — the user asked for a recommendation and the agent executed anyway;
- **scope expansion** — one bounded edit ballooned into a dependent workflow;
- **missing approval** — an approval was clearly required and the agent proceeded.

`AgentGovernancePlugin` (policy-as-code allow/deny) and the HITL approval gateway answer *"is this tool call permitted by policy?"*. Neither answers *"was the agent authorized to act at all, and within what scope?"* — and both require external infrastructure (a policy engine / an approval service).

## Proposal

Add `AuthorityRoutingPlugin`, a self-contained `BasePlugin` that assigns an **authority posture** to every tool call before it runs:

| Posture | Meaning |
| --- | --- |
| ADVISE | Recommendations only; no tool side effects. |
| EXECUTE | Act, within an explicitly bounded scope. |
| DEFER | Pause and request authorization when scope/approval is unclear. |
| STOP | Refuse when the work is not allowed. |

Two stages produce the final posture:

1. **Posture router** (optional, caller-supplied callable — typically a model call) returning a validated verdict with a scope statement.
2. **Deterministic guard** — an irreversibility keyword tripwire + an approval-state check — composed with the router under **most-restrictive-wins**: plain code can only make a posture *stricter*, never looser.

Design properties:

- **Fail-closed** — an unparseable or raised router verdict defaults to DEFER.
- **Guard-only mode** (no router) is fully deterministic and needs no model and no extra dependency.
- Complements, rather than duplicates, the existing governance plugins.

## Scope

- `src/google/adk_community/plugins/authority_routing_plugin.py` + export
- `tests/plugins/test_authority_routing_plugin.py` (deterministic, no credentials)
- `contributing/samples/authority_routing/` (runnable sample with live output)

Happy to adjust the API surface or placement to match maintainer preferences. I have an implementation ready and will open a PR referencing this issue.

Start by reading the existing BasePlugin and governance plugin behavior, then review the proposed paths in src/google/adk_community/plugins/authority_routing_plugin.py and tests/plugins/test_authority_routing_plugin.py. Add deterministic tests for posture routing, fail-closed handling, the irreversibility tripwire, approval checks, and most-restrictive-wins behavior, plus the runnable sample under contributing/samples/authority_routing/. Done means the plugin is exported and the sample produces live output without credentials.

Written by the indexing model from the issue text.