Improve Excerpt generation for non-english locales

[westi]

This is a follow on from #16079 to further improve the code.

While we improved it in 3.4 we can do a better job to more fully support the customisation for all different locales.

For known issues see the comments on the previous ticket.

[chriscct7]

This is something given the recent utf8mb4 that we should look at doing sooner rather than later