[timoklock]

Fixes #2442.

<string>.substring(start, end) returned <start> or <end> is greater than <string>.size() whenever end equalled the string's length and start < end, so 'abc'.substring(1, 3) failed instead of returning 'bc'. The strings extension defines the end offset as exclusive and only out-of-range or inverted bounds as errors, which is what cel-go implements.

Both SubstringImpl(…, start, end) overloads (string_view and Cord) return from inside the loop only when they reach end with input left. After the loop they handled only start == end == size. The post-loop check now records start and returns when end sits at the end of the input, the same two steps the loop body takes.

Tests:

• StringValueTest.Substring: Substring(4, 6), (0, 6) and (6, 6) on the six-code-point €€€€€€, for both the cord-backed and the view-backed value.
• StringsExtFunctionsTest (parse, check and evaluate): 'tacocat'.substring(4, 7) == 'cat', 'tacocat'.substring(0, 7) == 'tacocat', 'ta©o©αT'.substring(4, 7) == '©αT'.

Reference to test cases in cel-go

https://github.com/cel-expr/cel-go/blob/b6d796fb74036b0251ee055f824917cbf4ce6f18/ext/strings_test.go#L88-L93

[jnthntatum]

/gcbrun

[jnthntatum]

Thanks for the fix! LGTM -- this will get merged by the bot after internal approvals.