feat: support reference ranges in AnswerBuilder (e.g. [6-10])

[sjrl]

Summary

AnswerBuilder currently only handles individual document references (e.g. [1], [3]). It should also support ranges like [6-10], expanding them to the full set of indices {6, 7, 8, 9, 10}.

Current behaviour

_extract_reference_idxs uses re.findall with the user-supplied reference_pattern and calls int(idx) on each match. A range like [6-10] either doesn't match or fails to parse as an integer.

Expected behaviour

A reference of [6-10] should be treated as referencing documents 6, 7, 8, 9, and 10, equivalent to writing [6][7][8][9][10].

One Suggested implementation

Add an expand_reference_ranges: bool = False parameter to both __init__ and run(). It defaults to False for backwards compatibility. Otherwise users with custom reference_pattern values that legitimately capture strings containing - (e.g. "fig-3", "section-2") would otherwise get incorrect results silently.

_extract_reference_idxs gains the flag and handles both comma-separated parts and - ranges within each match:

@staticmethod
def _extract_reference_idxs(reply: str, reference_pattern: str, expand_ranges: bool = False) -> set[int]:
    matches = re.findall(reference_pattern, reply)
    idxs = set()
    for match in matches:
        if expand_ranges:
            # we split on comma to handle cases like `[1-3,7-9]`
            for part in match.split(","):
                part = part.strip()
                if "-" in part:
                    start, end = part.split("-", 1)
                    idxs.update(range(int(start) - 1, int(end)))
                else:
                    idxs.add(int(part) - 1)
        else:
            idxs.add(int(match) - 1)
    return idxs

When expand_reference_ranges=True the reference_pattern should also be updated to capture the broader form, e.g. \\[(\\d+(?:[,-]\\d+)*)\\].

---

👋 Hello there! This issue will be handled internally and isn't open for external contributions. If you'd like to contribute, please take a look at issues labeled contributions welcome or good first issue. We'd really appreciate it!