feat(py): Add client= parameter to Chat for batteries-included chatlas apps

[cpsievert]

Summary

Building a chatbot with shinychat + chatlas currently requires manually wiring up streaming, cancellation, and bookmarking — roughly 20-30 lines of boilerplate that look the same in every app. This PR adds a client= parameter to Chat that handles all of it automatically.

Express:

from chatlas import ChatAnthropic
from shinychat.express import Chat

chat = Chat(id="chat", client=ChatAnthropic())
chat.ui()

Core:

from chatlas import ChatAnthropic
from shiny import App, ui
from shinychat import Chat, chat_ui

app_ui = ui.page_fillable(
    chat_ui("chat", enable_cancel=True),
)

def server(input, output, session):
    chat = Chat("chat", client=ChatAnthropic())

app = App(app_ui, server)

No new API surface to learn — just one parameter on the class you already know. Client-specific methods live on a ChatClient wrapper at chat.client:

• .value — the raw chatlas client
• .set(new_client, sync=True) — swap models mid-session (deferred if streaming)
• .clear(client_history="clear") — reset UI and client turn history

In Express mode, enable_cancel defaults to True when a client is set.

R package changes

• chat_mod_server() now returns set_client(new_client, sync=TRUE) for swapping clients at runtime, and a status reactive ("idle" / "streaming").
• chat_restore() returns a cancel function for clean re-registration during client swaps.

Other changes

• enable_bookmarking() no longer raises when no bookmark store is configured — it silently no-ops instead.
• Docs lead with Chat(client=...) as the recommended path.

Test plan

• uv run pytest — unit tests verify ChatClient behavior (swap, clear, exports, messages_to_turns)
• uv run pytest pkg-py/tests/playwright — Playwright tests cover Express, Core, cancellation, and bookmark-no-store flows
• uv run pyright — type checking passes
• uv run ruff check pkg-py --config pyproject.toml — linting passes
• R package unaffected: cd pkg-r && Rscript -e "devtools::check(document=FALSE)" still passes

[cpsievert]

@gadenbuie there's a bit more review and polish I'd like to do here, but before I do, would be great to have a gut check from you in terms of the overall shape of the proposal here.