DuckDB Virtual Server

[texodus]

This PR adds a new Python module perspective.virtual_servers.duckdb, an implementation of the Perspective Virtual Server API (#3061) for DuckDB.

6mm rows from NYC Open Data:

DuckDB.Demo.mov

Using native DuckDB time_bucket function in perspective-viewer's Column Expression Editor to create time series bins:

DuckDB.Expressions.mov

• DuckDBHandler implementation of VirtualServerHandler for DuckDB in JavaScript and Python.
• VirtualServerHandler and VirtualServerSlice for JavaScript, Rust and Python.

Not implemented

This feature is still incomplete, but functional enough for testing. However, building Perspective from source is a daunting task, so as not to limit the potential audience of this brand-new feature, I've opted to go ahead and release it (core Perspective functionality is unaffected). Missing/broken functionality includes:

• split_by is "wonky" (especially with _ characters) due to not using recoverable alias column names.
• split_by does not preserve rollup aggregates.
• View::schema with a split_by does not behave the same as Perspective (does not de-normalize column group headers).
• kViewGetMinMaxReq unimplemented, so some column styling options that require bounds will not work.
• Only Columns & Rows serialization modes are supported.
• Duplicate data models - while host languages have completely different APIs for DuckDB clients available, there is a lot of duplicate code especialyl around SQL statement construction that could be shared in a common set of models.
• This is a SQL injection nightmare, and I'm not sure it's viable to fix this given the nature of the tool is arbitrary SQL exploration. Still great for local and WASM use!
• Inefficient SQL - most of the queries can be prepared, inefficient use of connections, etc.
• Some plugins could use some tuning around caching & batching behavior for extra-large datasets that DuckDB allows.
• API docs only (no user guide yet)