Next main job:
    redo parsing so as to properly handle full grammar and morphology,
    and ensure frees are correctly associated with their positions.

    I'm currently unsure whether this should be done by using a camxes
    implementation as an external program, or by implementing camxes
    ourselves.

    Once this is done, we can think about handling UI and so on.

fai after tagless jai is ignored; do we want that?

Multiple jai places:
    Currently we just swap with a single JaiPos, yeilding things like:

    > jai ca ke jai ri'a broda be fai do ke'e fai mi
    (ca)(). (ri'a)(mi). broda( )
    > ko'a jai ca jai bau jai ri'a broda fai ko'e
    (ca)({ko'a}). (bau)({ko'e}). (ri'a)(). broda( )

    Alternative would be to have a jai push a stack, and use subscripted fai
    to get at inner places.

xorxes suggests: pe [tag] ko'a = poi ke'a co'e [tag] ko'a
    (and that using xo'i is too specific)
    I'm tempted to agree.

Think again about zasni gerna tenses.

Support {sei sa'a} and {to'i}?

use to...toi in lojban for sides?

alternative tags mode, doing away with modals and non-logical sentence
connectives in favour of extra WithEventAs and TagRels?
    Nice idea, but I don't see how to handle complicated joik connected
    tenses; see notes/reductions.

option to render sumti (and tag, abstractor) connectives as quantifiers?

Pretty printing

optional unicode output

illocution: ko etc

irrealis UI - allow a free as a selbri tcita, then current free handling
should get the scope right, and we can consider them as modal operators?