fix: read ARRAY<DOUBLE> through Arrow's Float8Vector

[LuciferYang]

Fixes #1

Problem

The DoubleType branch of RowDataConverter.readArrayData cast the child vector to the private inner class Double8Vector — a wrapper that is never instantiated anywhere (its javadoc says "alias for Float8Vector"), a leftover from an unfinished refactor. Every read of an ARRAY<DOUBLE> column, via either Arrow representation that maps to it (List<Float64> and FixedSizeList<Float64>, i.e. Lance f64 vector columns), threw ClassCastException. The type layer accepts the schema (LanceTypeConverter documents FixedSizeList<Float64> <-> ARRAY<DOUBLE>), so DDL and planning succeed and the failure only surfaces at scan time. All read entry points (LanceSource, LanceInputFormat, LanceAggregateSource, LanceVectorSearch) funnel into this one private method and were all affected.

Fix

• Cast to org.apache.arrow.vector.Float8Vector directly, matching the sibling FloatType branch and the write path.
• Delete the dead Double8Vector wrapper.

Tests (RowDataConverterTest, 4 guards)

Test	Behavior it pins
testWriteThenReadArrayOfDoubleRoundTrip	List-representation write/read round trip: dense values, null element, empty array, null parent array
testReadFixedSizeListOfDouble	FixedSizeList (f64 vector column) read: values, null element, null parent, start-index arithmetic
testWriteBeyondInitialListCapacity	Writing 600 elements against an initial child capacity of 4 forces repeated setSafe reallocation — the path the production sink hits on every 1024-row batch (default write.batch-size)
testReadFixedSizeListOfFloat	The sibling FloatType branch (f32 vector columns, the primary embedding path): values and null element

Verification: 4/4 green; full module unit suite 234/0/0; six targeted mutations (reverting the cast fix, FSL start-index off-by-one, dropping the element null guard, dropping the parent-null short-circuit, setSafe -> set, wrong cast in the Float branch) each reddened exactly the corresponding test while the others stayed green; with the production fix reverted, the suite-level red set was exactly the tests above.

Build note

The new tests allocate off-heap Arrow memory, which requires --add-opens=java.base/java.nio=ALL-UNNAMED for MemoryUtil to initialize on JDK 17 (a CI matrix leg). The root pom now sets that argLine; jacoco appends its agent to the same property, so the two coexist.

[LuciferYang]

Superseded by upstream PR lance-format#69 (issue tracked at lance-format#68). Closing this fork-local PR.