Closing as all the technical arguments have been laid out, the rest is opinion.
Summary:
- I'm not convinced there are insurmountable difficulties with env vars, only inconveniences;
- I'm wary with ad hoc solutions to what should be handled by the build system, in this case cargo (is it actually a problem that needs to be solved? Maybe not, see previous point).
- This specific PR requires adding a file under this crate's directory, which basically requires using this crate as a patched dependency. Therefore it's only slightly cleaner than patching `build.rs` to hardcode `FFMPEG_DIR`. The solution is *less convenient* (and I'd wager rather confusing for users) than the tooling that was perceived as inconvenient in the first place, so not acceptable even as an ad hoc solution.