Editor: inline autocomplete (ghost text) from Translation Memory and Machine Translation

Look into offering inline, Tab-accepted completions in the translation editor, sourced from TM+MT suggestions already in memory for the current string, rather than querying on every keystroke.

Tab is currently key-bound to the dropdown autocomplete. Look into either conditionally showing ghost text on an empty dropdown, or, implementing the dropdown as ghost text (post feedback).

For the field being edited, take the text before the cursor. Against each candidate translation, find the longest common substring anchored at the end of the current editor content. Use the length of the match as a confidence signal and decide how much to offer based on that. The idea is to not propose a whole-sentence continuation and is to use match length as calibration:

• (terms below to be defined based on testing)
• match is short: suggest nothing
• match is longer: offer to complete the current word
• match is good, and the current word is already complete: offer to autocomplete the next word