Support optional DuckDB-backed inputs

[Copilot]

sccomp_estimate() could not dispatch on DuckDB-backed lazy tables because no matching S3 method existed. DuckDB should remain optional rather than becoming a hard package dependency.

• Optional dependencies: Add duckdb and dbplyr to Suggests.
• S3 dispatch: Register sccomp_estimate.tbl_duckdb_connection().
• Materialization: Check/install optional packages, collect the lazy table with dplyr::collect(), and delegate to the existing data-frame implementation.
• Regression coverage: Verify the DuckDB-specific S3 method is registered.

estimate <- sccomp_estimate(
  duckdb_table,
  formula_composition = ~ condition,
  sample = "sample",
  cell_group = "cell_group",
  abundance = "count"
)

• Fixes allow duckdb input without putting duckdb in imports but just in suggests, and using check_and_install #288

[stemangiola]

@copilot in unit test, convert the example data into duckdb and run estimate, and make sure the results are the same as with normal df

[Copilot]

@copilot in unit test, convert the example data into duckdb and run estimate, and make sure the results are the same as with normal df

Added a DuckDB regression test in commit 8dca580. It writes counts_obj to an in-memory DuckDB table, runs sccomp_estimate() through both paths, and compares the summarized estimates.

[stemangiola]

you should pass the dataset through until you reach the |> count(...) and then you can collect after the count

[Copilot]

Updated in commit fbe02f8: DuckDB-backed data now remains lazy through the existing sample/cell-group count() checks and is collected afterward, before the remaining data-frame modeling steps.

[stemangiola]

instead of collect_after_count, you can just check is(...) so you avoid creting too many arguments.

[Copilot]

Applied in commit 4bdacca: removed collect_after_count and now detect tbl_duckdb_connection directly in make_rectangular_data() before collecting.