fix(dynamodb): accept newline as UpdateExpression clause separator

[hampsterx]

Summary

Go AWS SDK v2 expression.Builder joins top-level UpdateExpression clauses with \n (e.g. "SET #a = :a\nADD #b :b"). Floci silently dropped clauses on these inputs because the clause-boundary parser only accepted a literal space as a word boundary. Fixes #430.

Three related fixes in DynamoDbService:

1. indexOfKeyword word-boundary relaxation. Accept any whitespace (space, tab, CR, LF) before a clause keyword via Character.isWhitespace, not just ' '. Before the fix, SET x = :a\nADD y :b lost both clauses: applySetClause greedily consumed ":a\nADD y :b" as the SET value, failed the placeholder lookup, and neither SET nor ADD ran.
2. indexOfKeyword loops past non-boundary hits. String.indexOf on the first non-boundary occurrence used to return -1 even when a later valid occurrence existed. An attribute named oldSET in "REMOVE oldSET SET newAttr = :v" masked the real SET. Loop until a boundary match is found or the string ends.
3. applyAddClause / applyDeleteClause / applyRemoveClause clause advancement alignment. Before this PR they advanced on the next comma before checking for a clause keyword, so an intra-clause comma in a following SET (e.g. "ADD a :v SET b = :b, c = :c") consumed the SET keyword. Aligned with applySetClause's existing "earlier of next clause keyword or next comma wins" rule.

Test cases added

All in DynamoDbServiceTest.java:

Case	Expression shape	What it exercises
1	SET #n = :n\nADD #c :inc	Newline between SET and ADD (primary repro)
2	ADD #c :inc\nSET #n = :n	Newline between ADD and SET
3	SET #n = :n\tADD #c :inc	Tab separator
4	SET #n = :n\r\nADD #c :inc	CRLF separator
5	SET #a = :a, #b = :b\nADD #c :inc\nDELETE #d :d	Canonical Go SDK three-clause shape
6	REMOVE #t\nSET #n = :n	REMOVE then SET across newline
7	DELETE #tag :d\nADD #c :inc	DELETE then ADD across newline
8	ADD #c :inc SET #n = :n, extra = :o	Intra-SET comma must not swallow SET keyword
9	REMOVE #t SET #n = :n, extra = :o	Same guard for REMOVE->SET
10	DELETE #tag :d SET #n = :n, extra = :o	Same guard for DELETE->SET
11	REMOVE oldSET SET newAttr = :v	indexOfKeyword loops past oldSET suffix
12	SET prefixSET = :v1, other = :v2	Attribute name containing SET not mis-parsed

Test plan

• ./mvnw test -Dtest=DynamoDbServiceTest (57 tests pass, 12 new)
• ./mvnw test full suite (2062 tests pass)
• Codex + Gemini agentic reviews on the diff (findings addressed or deliberately out of scope)

Out of scope

• The leading-keyword match in applyUpdateExpression (lines 892-904) still hardcodes a trailing literal space, so SET\t#a = :a at the start of an expression is still unsupported. No known SDK emits that shape and Go SDK only emits \n between clauses, not leading.
• The keyword constants passed to indexOfKeyword ("SET ", "REMOVE ", etc.) keep the trailing literal space. Switching to regex-based whitespace matching is a larger refactor and not needed to fix [BUG] DynamoDB UpdateExpression parser fails when clause types are separated by newline #430.
• Rewriting the hand-rolled parser with a formal grammar is a separate discussion.

Backwards compatibility

• Previously rejected inputs (newline/tab/CR-separated clauses) now apply correctly. No behaviour change for inputs the parser already accepted; all 45 pre-existing DynamoDbServiceTest cases still pass unchanged.

[hectorvent]

@hampsterx Thanks! This is a great addition to floci.