Normalise xsd:double text entry answers for Perseus derivation

[rtibblesbot]

Summary

Text entry answers like +5 and 1E3 are valid xsd:double but extract_value rejects them, so Perseus derivation emitted "answers": [].

• _perseus_number strips a leading + and lowercases E when the result parses as a finite extract_value number.
• _derive_text emits the normalised answer, or the original when none exists.
• INF, NaN and overflow (1e400) have no normalised form.
• +5% and +1/2 also normalise (to 5%, 1/2), since the check is extract_value, not xsd:double.
• extract_value not widened, to avoid changing legacy input_question behaviour.

References

Fixes #6224. Related: #6103, #6150, #6222.

Reviewer guidance

• Derivability gate (Older-Kolibri Perseus export of QTI Text entry items has no answers when the accepted answers are not numbers #6222, _all_numeric) is not on unstable; once merged it should call _perseus_number.

AI usage

Implemented with Claude Code from a pre-approved plan. Verified with TDD and manual review of the generated code.

🤖 Generated with Claude Code

Deviations from the issue spec

Criterion	Why not
Valid numeric answers (issue lists INF, NaN) derive a Perseus item	Perseus numeric-input and extract_value have no representation for them

---

@rtibblesbot's comments are generated by an LLM, and should be evaluated accordingly

How was this generated?

• Ran pre-flight CI checks (lint, format, tests) and verified all pass
• Rebased onto the target branch and resolved any conflicts
• Reorganized commit history into clean, logical commits
• Audited the diff to ensure only issue-relevant files are changed
• Built PR body from the repository's PR template with evidence blocks

🟡 Waiting for feedback Last updated: 2026-10-02 22:23 UTC

[rtibblesbot]

Outside this PR's scope — tracked under #5998:

• Older-Kolibri Perseus export of a QTI Text entry item answered 1_000 or ٥ has no answers #6281 Older-Kolibri Perseus export of a QTI Text entry item answered 1_000 or ٥ has no answers

---

@rtibblesbot's comments are generated by an LLM, and should be evaluated accordingly

How was this generated?

• Ran pre-flight CI checks (lint, format, tests) and verified all pass
• Rebased onto the target branch and resolved any conflicts
• Reorganized commit history into clean, logical commits
• Audited the diff to ensure only issue-relevant files are changed
• Built PR body from the repository's PR template with evidence blocks

[rtibbles]

Clear test cases - simple normalization code fix.