fix: audit array_insert expression for correctness and test coverage

[andygrove]

Which issue does this PR close?

N/A - proactive audit of existing expression.

Rationale for this change

Audited the array_insert expression against Spark 3.4.3, 3.5.8, and 4.0.1 source code to verify correctness and identify test coverage gaps. Found a metadata bug and significant missing test coverage.

What changes are included in this PR?

Bug fix:

• Fixed nullable() in ArrayInsert (Rust) to account for pos_expr nullability, matching Spark's first.nullable | second.nullable. Previously only checked src_array_expr, so if the position column was nullable but the array was not, the metadata incorrectly reported "not nullable".

Test coverage (expanded from 1 query to 27+ queries):

• Column and literal arguments for int, string, boolean, double, float, long, short, and byte array types
• NULL handling: null array, null position, null value, arrays with existing nulls
• Positive out-of-bounds insertion with null padding (pos=5, pos=10)
• Negative indices: in-range (-1 through -4) and out-of-bounds (-5, -10) with null padding
• Special double values: NaN, Infinity, -Infinity, negative zero
• Multibyte UTF-8 characters and empty strings
• Legacy negative index mode (spark.sql.legacy.negativeIndexInArrayInsert=true) in a separate test file
• Parquet dictionary encoding via ConfigMatrix

Documentation:

• Added docs/source/contributor-guide/expression-audit-log.md to track audited expressions
• Updated audit skill with Step 8 to maintain the audit log

How are these changes tested?

All new SQL file tests pass against both Spark and Comet:

./mvnw test -Dsuites="org.apache.comet.CometSqlFileTestSuite array_insert" -Dtest=none

Rust unit tests and clippy pass. Full build succeeds.

[andygrove]

@kazuyukitanimura @martin-g This PR was created using the audit skill that you reviewed

[andygrove]

I think it is important to keep track of which expressions have been audited and against which Spark versions. Maintaining a markdown doc is one approach. Another approach could be to add comments and/or annotations directly in the code in the serde handlers.

[martin-g]

@kazuyukitanimura @martin-g This PR was created using the audit skill that you reviewed

Really cool!