Comma fused via underscore into an MWE token has a space before it, not after (Opelika_,_Ala.)

Follow-up to #57's genitive \`'s\` case and the same general class of bug: when a comma gets fused via underscore into a multi-word proper-noun/title token (\`Opelika_,_Ala.\`, \`Boulder_,_Colorado\`, \`Hallmark_Cards_,_Inc.\`), the underscore *before* the comma is wrong -- a comma is always flush with the word before it and followed by a space, never the reverse (\`Opelika, Ala.\`, not \`Opelika , Ala.\`). Confirmed directly against \`src/semcor/brown-nolines.txt\` for every instance below.

31 raw candidates found (tokens containing \`_,_\`); **20 confirmed and fixed** (pure single-token \`_,_\` -> \`,_\` substitution, no merge, no sense/lemma impact). 11 left for later, separate work because fixing the comma alone wouldn't make them match the reference either way:

- 6 also embed a numeric range that needs its own hyphen fixed flush (\`400_-_401\` -> Brown has \`400-401\`, no spaces at all) -- the same shape as #9's number-range hyphens, just inside one of these MWE tokens instead of separate tokens.
- 4 are Selective Service classification codes (\`4_,_-_D\`, \`1_,_-_A\`) using an entirely different escape convention in the reference (\`/4,-~D\`) that needs its own investigation.
- 1 (\`Norman_B._Small_,_Jr.\`) is missing a comma entirely compared to Brown's real \`Norman B., Small, Jr.\` -- a content gap, not a spacing bug.

Start with src/semcor/brown-nolines.txt and inspect the 31 tokens containing `_,_`, using the issue's confirmed cases and Brown reference as the guide. Change only the 20 pure single-token cases to `,_`; leave the 11 cases requiring separate range, classification-code, or content-gap work untouched.

Written by the indexing model from the issue text.