[common] Fix NullPointerExceptions on null transform inputs

[plusplusjiajia]

Purpose

Three NullPointerExceptions on null transform inputs, all reachable today:

• TrimTransform reads charsToTrim with toString() before checking it for null, so a null second input throws — even though StringUtils.ltrim/rtrim, which it delegates to, already define a null charsToTrim as a null result.
• SubstringTransform reads a begin or length field with InternalRow.getInt without an isNullAt check. On a GenericRow that throws; on a columnar row it returns an undefined value, so the rule masks at an arbitrary offset without failing.
• StringTransform.toString maps inputs with Object::toString and throws on a null input, which ConcatTransform and ConcatWsTransform accept.

All three change behaviour only for inputs that previously threw, so no query that previously succeeded changes its result, and nothing about the persisted JSON format changes.

[JingsongLi]

SUBSTRING('123', null) still throws an NPE at toString() and does not follow SQL null propagation; consistently reproducible locally.

[plusplusjiajia]

SUBSTRING('123', null) still throws an NPE at toString() and does not follow SQL null propagation; consistently reproducible locally.

@JingsongLi Thanks, fixed.

The old code only checked for null when the position was a FieldRef. A literal null went straight to Integer.parseInt(...) and threw. Now any null input gives null, wherever it comes from:

SUBSTRING('123', null) -> null
SUBSTRING('123', 1, null) -> null

One change worth noting: the null check now runs before the "begin past the end" check, so SUBSTRING('123', 99, null) returns null where it used to return ''. That keeps one simple rule — null always wins — and matches Spark, whose Substring is NullIntolerant.