Fix dead regex alternation: mm/MM unreachable in unit pattern

[MaxwellCalkin]

Summary

• The regex unit group in extract_numbers.py has [TBMKtbmk]n? before mm|MM. Since m is in the character class [TBMKtbmk], a single m matches first when the input is "mm", leaving the second m unconsumed. The mm|MM alternation is effectively dead code — it can never match.
• This means financial values like "$500mm" (a common abbreviation for millions in finance) are captured with unit m instead of mm, which happens to still normalize correctly only by luck — but the captured unit string is wrong, and downstream logic relying on the exact unit value would break.
• Fix: Move mm|MM before the single-character class so the two-character token gets regex priority.

Test plan

• Verify that "$500mm" is captured with unit mm (not m)
• Verify that "$500MM" is captured with unit MM (not M)
• Verify that "$500M" still captures unit M correctly
• Verify that "$500Mn" still captures unit Mn correctly

🤖 Generated with Claude Code

AI Disclosure

This PR was authored by Claude Opus 4.6 (Anthropic), an AI agent operated by Maxwell Calkin (@MaxwellCalkin).