Add managed import modes for CSV and Parquet ingestion

[daniel-thom]

Summary

Add managed import modes so users can choose whether discovered CSV and Parquet files are mounted as source-backed views or materialized into DuckDB tables.

This should improve performance for CSV-heavy projects and make Parquet-backed projects easier to transport as a single DuckDB-based datasight project.

Problem

Today, users who start from flat files do not have an explicit, inspectable ingestion mode:

• CSV files often perform better when imported into DuckDB tables instead of being queried through repeated scans.
• Parquet files are portable, but source-backed references make project sharing and relocation more fragile.
• The current project setup does not make the tradeoff between view and table visible enough.

Proposed scope

• Add an ingestion mode concept with auto, view, and table.
• Expose this in the CLI for project setup and file-add flows.
• Persist import metadata so users can tell how a table/view was created and from which source.
• Materialize managed tables into the project DuckDB database with deterministic naming.
• Make auto prefer table for CSV and view for Parquet unless a later design decision changes that default.

CLI sketch

datasight init --import-mode auto
datasight add generation.csv --import-mode table --table-name generation_fuel
datasight add data/*.parquet --import-mode view

Acceptance criteria

• Users can choose auto|view|table during relevant CLI ingestion flows.
• CSV and Parquet ingestion paths both support managed-table materialization.
• Imported objects carry source metadata that can be surfaced later in the UI and diagnostics.
• The chosen import mode is visible to the user after ingest.
• Tests cover CSV and Parquet imports for each supported mode.

Notes

• This is the lowest-risk entry point into the broader “data prep and portability” roadmap.
• Follow-on UI affordances can read from the same import metadata.

[daniel-thom]

Implemented in #53