Not using semicolons seems fine enough. Going through and removing the semicolons that you already used seems like you're just trying to start drama. Why not use that time to do something that effects the code positively or negatively?
FWIW, it took about 10 minutes to remove the semicolons from the codebase, with a simple grep: /;$/
For the few places where a ; is actually needed you can run a regexp based search as well, the rest is easily identified by unit tests.
We've also spend some 50+ hours on writing new code, tests, a new build and automated tests system, a completely new documentation site and answering countless issues on GitHub.
So yes, you could say my taste changed, but there's good a good reason behind it—the semicolons are optional, and Zepto is all about concise code.
A correct minifier works on the AST of the code anyway (at least in the abstract). It works on the semantic meaning of the code, so it interprets semicolons or the lack thereof exactly the way any other implementation would.
The first case with the broken inlining tool is typically not actually an issue, because your semicolonless code gets put through a minifier which probably adds a bunch of semicolons again :)
And honestly this is, intentionally, a tiny library. If it took more then 10 minutes I'd be surprised.