test: add unit tests for utility classes

[LuciferYang]

Summary

• Add unit tests for BlobUtils, LargeVarCharUtils, Utils, and VectorUtils covering all reachable branches in both Spark and Arrow field-checking methods
• Tests cover null inputs, missing/wrong metadata, case-insensitive value matching, type-check isolation (non-String for LargeVarChar, non-numeric for Vector), and edge cases like empty version lists and malformed metadata
• UtilsTest helper builds ZonedDateTime directly instead of re-implementing the production micros-to-Instant conversion, keeping test logic independent from source logic

Test coverage details

Class	Methods tested	Tests
BlobUtils	isBlobSparkField, isBlobArrowField	12
LargeVarCharUtils	isLargeVarCharSparkField, isLargeVarCharArrowField, createPropertyKey	14
Utils	parseVersion, findVersion	11
VectorUtils	isVectorField, getVectorDimension, isVectorArrowField, getVectorArrowDimension, shouldBeFixedSizeList, createVectorSizePropertyKey	25

Test plan

• mvn test -pl lance-spark-base_2.12 -Dtest="org.lance.spark.utils.*Test" — 108 tests pass

[hamersaw]

It looks like the majority of these tests are essentially "x=10; assertTrue(isXTen(x))". It doesn't feel like there is a great deal of added value, the regressions would occur if upstream began setting X to something else, which we really can't test. I'm wondering if there's a better way to detect these kinds of issues.

[LuciferYang]

@hamersaw Fair point — pushed 5b8bb14 trimming this:

• Dropped the *Valid tests from BlobUtilsTest and LargeVarCharUtilsTest — they were the literal "x=KEY; assertTrue(isKey(x))" tautology you called out. The CaseInsensitive cases already cover the positive path with non-tautological input (and would catch an accidental .equals regression).
• Replaced constant references in test inputs with string literals ("lance-encoding:blob", "arrow:large-var-char", "arrow.fixed-size-list.size"). A rename of the production constant now breaks the test — so the wire-format contract is pinned on the test side, which it wasn't before.
• Fixed testCreateVectorSizePropertyKey the same way (was concatenating the constant into the expected value).

58 tests total, all pass locally. Javadoc on each class now notes that upstream-convention drift is the integration layer's job — BaseBlobCreateTableTest, SchemaConverterTest, and the LanceArrow* suites already round-trip through real lance data and would catch that.

Let me know if this is the direction you had in mind, or if you'd rather drop the unit tests entirely and lean on the integration coverage.