Unstructured-inference

* refactor: add a class `ElementType` for the element type constants and use the constants to replace element type strings
* enhancement: support extracting elements with types `Picture` and `Figure`
* fix: update logger in table initalization where the logger info was not showing
* chore: supress UserWarning about specified model providers

* change the default model to yolox, as table output appears to be better and speed is similar to `yolox_quantized`

* chore: remove logger info for chipper since its private
* fix: update broken slack invite link in chipper logger info
* enhancement: Improve error message when images extracted doesn't match page layouts.
* fix: use automatic mixed precision on GPU for Chipper
* fix: chipper Table elements now match other layout models' Table element format: html representation is stored in `text_as_html` attribute and `text` attribute stores text without html tags

* Handle kwargs explicitly when needed, suppress otherwise
* fix: Reduce Chipper memory consumption on x86_64 cpus
* fix: Skips ordering elements coming from Chipper
* fix: After refactoring to introduce Chipper, annotate() wasn't able to show text with extra info from elements, this is fixed now.
* feat: add table cell and dataframe output formats to table transformer's `run_prediction` call
* breaking change: function `unstructured_inference.models.tables.recognize` no longer takes `out_html` parameter and it now only returns table cell data format (lists of dictionaries)

* Allow table model to accept optional OCR tokens

* Fix: include onnx as base dependency.