[Bug] Japanese iPhone keyboard fails to flush the suggestion queue

Severity level: Slight annoyance. Once you get used to it, you can avoid it easily, though occasionally you forget to do the special thing to avoid it, and you send the same text twice.

Reproduce:
- Open iOS build with Japanese ten-key keyboard open. (Called "Japanese - Kana")
- Open a chat and start typing anything. (hitting 3 random keys should queue up the first letter of that key and give you Chinese symbol choices) Then hit the lower right key to "lock in" the letters. Then type three more random letters but don't lock them in and don't tap any of the suggestions.
- With the characters queued and the suggestions displayed, hit the "Send" button on the Keybase UI.
- Notice that it only sends the characters that were locked in. AND as a side effect, the entire message is still in the entry box waiting for lock in.

Expected:
- Doing the same on slack will auto-lock-in whatever you have queued up and then send it. (Try the above in slack)

It's a slight annoyance... but we see chats all the time like:

"Hey, about the bug yesterday"
"Hey, about the bug yesterday, I fixed it and put up a pull request"

Some people never lock-in and write full sentences, some people lock in frequently so you get two of the same message with just an extra word at the end... it's all over the place...

but yeah, we've gotten around by saying to always lock in before sending on keybase.

Sorry for the inconvenience. Japanese and the way it's IME work are hard to code for, and knowing you guys aren't even officially supporting us makes us feel even more sorry (lol)

Please let us know if you can fix this issue. If you want a video, I can capture one.

Start by reproducing the Japanese - Kana keyboard behavior on the iOS build, using the chat composer and Send button as the entry points. Trace how queued IME text is handled when sending, then verify that sending includes the queued characters and clears the entry box.

Written by the indexing model from the issue text.