* README: add note on JIT vs AOT perf

True story. Observed on osm2rdf data, when converting ~7M triples from Turtle to Jelly. With JIT it took ~7s, while with AOT it was ~25–30s, depending on how the stars were aligning. Tested on the same machine as listed in Jelly benchmarks on the website.

The JIT throughput would be in the ballpark of 1MT/s, which is what we would expect from the Turtle parser (it tops out around this point). The AOT throughput is just abysmal.

* fix

* fix2

* Move awkward field iteration inside FrameInfo

* Add all details functionality

* Add grouping functionality and fix frame aggregation

* Add some documentation

* Add complex jelly file tests

* Apply suggestions from code review

Co-authored-by: Piotr Sowiński <ostrzycielnozyczek@gmail.com>

* Remove unused field

* Rename 'flat' to 'all' and make its output nested

* Add 0's filtering from output

* Make nested results ordered

---------

Co-authored-by: Piotr Sowiński <ostrzycielnozyczek@gmail.com>

Does the same thing, cleaner syntax.

Closes #74

Everything came together really nicely with context parameters.

One thing is that I'm not sure how accurate the statistics are. I use getSerializedSize() of ProtoMessage as the source (and length of encoded string + 1 for the tag byte), but I'm not sure if these are correct.

I was hoping that triples_size would equal subject_size + predicate_size + object_size, but that isn't the case.

…164)

* Initial generalized rdf inference implementation

* Add jelly-text test for options

* Mark NTriples and NQuads as supporting generalized statements