Normalize spoken numbers and times in cleaned text

[frarredondo]

Summary

ASR often writes times and numbers in awkward spoken form that hurts readability.

Motivation / evidence

Examples from cleaned transcripts:

• seven point zero AM
• 6 07 a.m.
• similar patterns for versions, money, and counts

Proposal

Extend cleanup (or a dedicated normalization pass) to convert common spoken times/numbers into conventional written form, behind --cleanup or a more specific flag.

Out of scope

Full numeric NLP for every locale; changing timed segment text unless we decide on a parallel normalized field.

Acceptance criteria

• Common time patterns in examples normalize cleanly
• Conservative rules with unit tests for false-positive cases
• Docs / --help updated if user-facing