feat: add QueryResult::data_str to read results without copying

[austin-barrington]

Right now if you want a query result as text you've got data_utf8(), which copies the whole result buffer into a new String, or data_utf8_lossy(). There's no way to just look at it as a &str.

This adds QueryResult::data_str() -> Result<&str>. It checks the bytes are valid UTF-8 in place and borrows straight from the result buffer, so nothing gets allocated.

On a 92MB JSONEachRow result (3M rows):

• data_utf8(): ~45ms, plus a 92MB allocation
• data_str(): ~6ms, no allocation

Also:

• New Error::InvalidUtf8 for when the data isn't valid UTF-8 (binary formats like RowBinary or Parquet). Error is #[non_exhaustive] so it's not a breaking change.
• data_utf8() is unchanged, its docs now point at data_str() if you don't need an owned string.
• examples/03_query_results.rs and docs/examples.md show it.

Tests cover a normal result (and check it points at the same memory as data_ref()), a binary result returning InvalidUtf8, and an empty result.

Note

Add QueryResult::data_str to return result text as a borrowed &str

• Adds a public data_str method that validates the result buffer in place with std::str::from_utf8 and returns a borrowed string slice without allocating. See query_result.rs.
• Adds an Error::InvalidUtf8 variant carrying std::str::Utf8Error for results that are not valid UTF-8, such as binary output. See error.rs.
• Updates docs and the main example to show the borrowed read alongside the existing owned, lossy, and byte-slice accessors.
• Adds tests for zero-copy borrowing, binary rejection, and empty results in query_result.rs.

^{Macroscope summarized 085dcd2.}