feat(workspaces): mount volumes and name the container of a DockerWorkspace

[DEENUU1]

Problem

pydantic-deep's CLI runs in Docker with the project mounted at /workspace and, for a named workspace, the same container every time it starts. Moving it onto Pydantic AI workspaces (the follow-up to #112) needs both from DockerWorkspace, which exposed neither. They were reachable only by building DockerWorkspaceBackend with a sandbox factory of your own.

Separately, the package shipped no py.typed, so type checkers in projects that install it read every name as Any. pydantic-deep's mypy run surfaced it.

Change

• DockerWorkspace(volumes={host: container}) passes the mounts to DockerSandbox.
• DockerWorkspace(container_name=...) / DockerWorkspaceBackend(container_name=...): one container named by whoever configures the workspace. It is created on first use and attached after, by this process or the next. A ref naming it must still find it there (WorkspaceUnavailableError otherwise), and a ref naming any other container is declined by the capability and refused by the backend. destroy accepts it.
• src/pydantic_ai_backends/py.typed.

Verification

• 1320 tests at 100% coverage; ruff, pyright and mypy are clean.
• New unit tests cover: creation under the configured name, attaching while the container exists, an unavailable ref once it is gone, a foreign ref refused and declined, destroy, and volumes reaching the sandbox.
• On a real Docker daemon: a host file is readable inside the container, a file written inside appears on the host, and a second backend reaches the same container by name.
• The built wheel contains pydantic_ai_backends/py.typed.