fix: do not split "B"-encoded-words at UTF-8 char boundaries

[link2xt]

Fixes #40

[link2xt]

I also tested that it fixes the problem in chatmail/core#7039

[link2xt]

@mdecimus Could you review this? This is the remaining piece of fixing the issue with long unicode group names in Delta Chat: chatmail/core#7039

[mdecimus]

There is some code duplication here, =?utf-8?B? is being written here and also on line 47.
I prefer an approach using an iterator that returns the bytes at their correct offest which are then encoded and wrapped around =?utf-8?B? and ?=.
In fact, I don't think that a custom iterator is needed at all, you could use char_indices (and avoid having to do (ch as i8) >= -0x40), keep track of the offsets and then print the encoded word once you reached the right size.

[link2xt]

I force-pushed the change, now using char_indices.

[mdecimus]

I made some changes to the B-encoding function, can you check that it works for you so I can release?

[link2xt]

I tested with commit 2ec9d02, it works.

Why remove debug_assert though?

-        // There is always a header or continuation whitespace before inline text.
-        debug_assert!(bytes_written > 0);
-

If the function is called with bytes_written equal to 0, it starts the header with \t. This should not happen, so debug_assert ensured that this kind of calls will not be introduced in the future.