rpkg: dvs_init returns config as a single-row tibble

[CGMossa]

AI-written details

Summary

• dvs_init previously returned list(status = "initialized") and printed DVS Initialized to stderr, giving R callers no programmatic view of what was actually configured.
• Now returns a one-row tibble:

  > dvs_init(storage_path = "/tmp/storage", compression = "none",
  +          metadata_folder_name = "custom_meta", group = "staff")
  # A tibble: 1 x 4
    compression metadata_folder_name backend_path  backend_group
    <chr>       <chr>                <chr>         <chr>
  1 none        custom_meta          /tmp/storage  staff

• Return type changes from Result<List> to Result<ColumnarDataFrame>; builds with vec_to_dataframe(&[config])?. No view struct: Backend already has #[serde(untagged)], so Backend::Local(LocalBackend{..}) flattens transparently into backend_path / backend_group via miniextendr's recursive nested-struct flattening. cli: Option<CliConfig> is skipped at init time by its existing skip_serializing_if = "Option::is_none".
• Drops the invisible flag and the r_println!("DVS Initialized") line. The printed tibble is now the visible signal.
• R wrapper wraps the result in tibble::as_tibble() for parity with dvs_add / dvs_get / dvs_status.
• Adds a testthat case asserting shape, column names, and values.
• Adds ui/main_init.sh exercising the option matrix (defaults, --no-compression, custom metadata folder, group, all combined, storage-inside-repo guard) so the CLI dvs.toml can be eyeballed alongside the R tibble.
• justfile includes init in ui_names so just ui-publish picks it up.

Carved out: ui-trace cleanup (xtrace filter + deferred set -x) moved to #198.

Test plan

• just rpkg-test passes (cargo + testthat, including the new init tibble test)
• bash ui/main_init.sh runs to completion across all 6 scenarios; CLI dvs.toml matches R tibble values

Notes

• When metadata_folder_name is left at its default (None), it appears as a list-column with a single NULL rather than NA_character_, because miniextendr's columnar serializer cannot infer the inner type from a bare None. With any explicit string it's a clean <chr> column. Functional but not pretty; left as follow-up.

---

Drafted by Claude (claude-opus-4-7). Reviewed by the author.

[dpastoor]

like it!

[CGMossa]

This is a good start, but I would like to try and resolve A2-ai/miniextendr#307 before going further with this. The Option<..> fields become list-cols, even though they ought not to. This is a limitation due to the serde serialization model, and R's nuanced/typed NA. I would like these inits to be stackable, and as list-cols, they are not, if in another invocation an actual value is passed.