fix(r): make table_name optional for deferred QueryChat$new(NULL)

[cpsievert]

The problem

QueryChat$new(NULL) — constructing without a data source, to register one later per Shiny session via $server(data_source = ) — currently requires table_name up front too:

QueryChat$new(NULL)
#> Error: `table_name` is required when `data_source` is `NULL`.

That's stricter than it needs to be. The only thing table_name is used for in this deferred case is computing a default Shiny module ID (querychat_<table_name>) so $ui()/$sidebar() — which have to be built once, before any session (and therefore any real table) exists — get a readable, collision-free namespace when an app has more than one QueryChat instance. For a single-instance app, a generic default ID works fine, and Python's QueryChat() already allows omitting it for exactly that reason.

The fix

• table_name is now optional when data_source is NULL. If omitted (or explicitly NULL), $id falls back to the generic "querychat", matching Python's behavior. You can still pass id = yourself if you want a custom one.

• $server() gains a table_name parameter, so a name can be supplied per-session instead of only at construction time:

  qc <- QueryChat$new(NULL)  # no table_name needed just to build the UI
  
  server <- function(input, output, session) {
    conn <- get_per_user_connection(session)
    qc$server(data_source = conn, table_name = "orders")
  }

• If data_source is given to $server() and no name can be resolved from any of $server(table_name = ), $new(table_name = ), or an already-registered table, you now get a clear error instead of a cryptic base-R "subscript out of bounds".

• As part of this, $add_table()'s single-table id auto-rename (rewrite $id to querychat_<table> when one table is registered) is removed rather than worked around: the id is now fixed at construction time, matching Python. The rename was nearly unreachable — $remove_table() refuses to remove the last table, and the constructor registers inline — and in its one live path (registering a table between $ui() and $server()) it could itself desync the module namespace from the already-rendered UI. This replaces an earlier revision of this PR that pinned the id via a private .id_pinned flag.

Test plan

• Updated/added tests in test-QueryChat.R: QueryChat$new(NULL) no longer errors and falls back to the generic ID; explicit table_name = NULL behaves the same as omitting it; $server(data_source = ) gives a clear error when no name can be inferred; $server(data_source = , table_name = ) registers under the given name.
• make r-check-format, make r-check-tests (2083 passed), and make r-check-package (0 errors, 0 warnings; the one pre-existing NOTE is unrelated to this change) all pass.

Related to #300, but independent of it — this is about deferred-construction ergonomics, not the concurrent-session safety issues that issue tracks.

[Copilot]

🟢 Approval recommended

No unresolved issues blocking approval were identified.

Review details

Files not reviewed (1)

• pkg-r/man/QueryChat.Rd: Generated file

• Files reviewed: 3/4 changed files
• Comments generated: 0 new
• Review effort level: Lite