fix: bound unclosed comment source positions to input

[feritcemkahraman]

Parsing an unfinished comment such as /*a puts its source end offset at 4 even though the original input has only 3 characters. The root end is also beyond EOF. This happens because the fault-tolerant tokenizer can advance past the input and Input.fromOffset() returns a location even for that offset.

Clamp the comment end to the last original character and the root end to EOF. Preserve the usual convention: comment line/column identify its last character, while the offset is exclusive. Add a multiline unfinished-comment regression covering both source ranges and extraction of the original comment text.

Validation: the new regression fails before the fix; all 49 unit tests pass after it. npm run test:lint and npm run test:integration pass (33 real-world resources).

[ai]

Thanks