Inaccurate license detection based on lgpl_bare_single_word.RULE

A recent scan of MANIFEST.MF in jgrapht-core-0.9.1.jar returns [lgpl-2.0-plus](https://scancode-licensedb.aboutcode.org/lgpl-2.0-plus) as the license based on the clue "LGPL" in the text. That makes sense if that's all you have, but the actual text in the original MANIFEST.MF file is actually more specific:

line 37
`Bundle-License: http://jgrapht.sourceforge.net/LGPL.html`
and if you go to the link you see that it specifically identifies lgpl-2.1

Not a major problem, but returning lgpl-2.1 would be better and more accurate. Perhaps a new detection RULE would address this.

[MANIFEST.MF.zip](https://github.com/nexB/scancode-toolkit/files/11366628/MANIFEST.MF.zip)