[OsamaAnsar]

Description of change

Two cases where validateObjectAgainstGbnfSchema (used by LlamaJsonSchemaGrammar.parse and function-call param parsing) accepts values that the grammar for the same schema can never produce:

1. timeRegex / datetimeRegex had an optional hour in the timezone offset ([+-]([01][0-9]|2[0-3])?:[0-5][0-9]), so 12:00:00+:30 and 2024-01-01T10:00:00-:00 were valid. The grammar in GbnfFormatString.ts requires the hour (("+" | "-") ([01] [0-9] | "2" [0-3]) ":" [0-5] [0-9]). The ? is removed.
2. validateObject checked typeof object !== "object" || object === null, so an array passed for an object schema (typeof [] === "object"). An object schema's grammar starts with {, so an array is never valid. Arrays are now rejected, with the error message saying "array" instead of "object" (and "null" for null).

Before / after (real output from validateObjectAgainstGbnfSchema, run with npx tsx; "throws" is LlamaJsonSchemaValidationError):

                                            before    after
time "12:00:00+:30"                         true      throws: Expected a valid time string but got "12:00:00+:30"
date-time "2024-01-01T10:00:00-:00"         true      throws: Expected a valid date-time string but got ...
time "12:00:00+05:30" / "12:00:00Z"         true      true
[] vs {type: "object", properties: {}}      true      throws: Expected an object but got "array"
{} vs {type: "object", properties: {}}      true      true
[] vs oneOf [object, null]                  true      throws: Expected one of 2 schemas but got []

Tests: new standalone test/standalone/utils/validateObjectAgainstGbnfSchema.test.ts (does not need the native addon or a model); it also asserts that the grammar text for the schema requires the offset hour and that an object grammar starts with {. With only the source fix reverted 5 of the 10 tests fail; with the fix all 10 pass. tsc --noEmit and eslint on src/utils/gbnfJson and the new test pass.

Not tested: the native addon could not be built in my environment, so LlamaGrammar.test.ts and other suites that need it, and the model-dependent tests, were not run; the grammar was checked only as text, not compiled by llama.cpp.

Pull-Request Checklist

• Code is up-to-date with the master branch
• npm run format to apply eslint formatting
• npm run test passes with this change - the new tests, tsc --noEmit and eslint pass; tests that need the native addon / models were not run
• This pull request links relevant issues as Fixes #0000 - N/A, no existing issue found
• There are new or updated unit tests validating the change
• Documentation has been updated to reflect this change - N/A, no API or usage change
• The new commits and pull request title follow conventions explained in pull request guidelines (PRs that do not follow this convention will not be merged)