feat(pkg-r): support sqlite engine in PinSource

[gadenbuie]

Summary

PinSource previously always used DuckDB, even for pins that deserialize to a plain data frame. Since duckdb and RSQLite are both optional (Suggests) in the R package, a user who installed only RSQLite could use DataFrameSource but not PinSource. This PR aligns PinSource with DataFrameSource's engine handling.

Key changes:

• Extracted DataFrameSource's inline connection setup into a shared internal helper, new_dataframe_connection(df, table_name, engine), which creates an in-memory connection, registers the data frame, and applies the DuckDB security lockdown. DataFrameSource$initialize() now calls it.
• PinSource$new() gains an engine argument (default getOption("querychat.DataFrameSource.engine", NULL), resolved via the same get_default_dataframe_engine() as DataFrameSource). Data-frame pins now share DataFrameSource's code path, so the SQL flavor reported by the inherited get_db_type() is resolved consistently.
• The efficient direct DuckDB file read is kept for parquet/CSV/JSON pins under the duckdb engine. Requesting engine = "sqlite" for those pin types falls back to pin_read() with a warning that DuckDB reads them more efficiently.
• duckdb is no longer unconditionally required by PinSource — it is only needed when actually used.

While here, this PR also backfills a changelog entry for the original PinSource feature (#246), which was merged without one, in both NEWS.md and the Python CHANGELOG.md. The Python side is otherwise intentionally unchanged: duckdb is a core dependency there and DataFrameSource/PinSource are DuckDB-only by design, so no equivalent gap exists.

Verification

board <- pins::board_temp()
pins::pin_write(board, mtcars, "mtcars", type = "rds")

# Force the SQLite engine for a data-frame pin
ps <- querychat::PinSource$new(board, "mtcars", engine = "sqlite")
ps$get_db_type()
#> [1] "SQLite"
ps$execute_query("SELECT * FROM mtcars WHERE mpg > 30")

# A file-type pin under sqlite warns, then reads via pin_read()
pins::pin_write(board, mtcars, "cars", type = "csv")
ps2 <- querychat::PinSource$new(board, "cars", engine = "sqlite")
#> Warning: Reading csv pin "cars" into SQLite.
#> i The duckdb engine reads csv pins more efficiently. ...
ps2$cleanup()
ps$cleanup()

New tests cover the SQLite data-frame path and the file-type fallback warning in tests/testthat/test-PinSource.R.

[gadenbuie]

Python E2E failures are unrelated, this PR only touches R code