[Experimental] Zenzai (#92)

* experimental rinna integration

* Update impl

* update

* Bump swift-actions/setup-swift from 1 to 2

Bumps [swift-actions/setup-swift](https://github.com/swift-actions/setup-swift) from 1 to 2.
- [Release notes](https://github.com/swift-actions/setup-swift/releases)
- [Commits](https://github.com/swift-actions/setup-swift/compare/v1...v2)

---
updated-dependencies:
- dependency-name: swift-actions/setup-swift
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>

* change test

* change impl

* take gpt2 weight as option

* don't use async

* support status check

* enhance error

* avoid percent encode

* update

* GPT-2 based kana-kanji conversion is now perfectly workinggit statusgit status

* fix a bug

* Rename gpt2/llama -> zenz

* cleanup

* internal apiを綺麗にした

* cleanup experimental commands

* update

* partially support incremental input using cache

* fix names

* fix bug

* support roman2kana

* cleanup

* fix minor bugs

* improve logic

* fix minor bug

* fix minor bug

* fix minor bug

* optimize

* optimize performance

* Optimize cache hit

* cli: add anco session command

* fix cache hit bugs

* improve session commands

* maybe this will work better for incremental input environment

* speed up zenzai by using n_best alternatives

* update zenz context

* adding no_typo api

* add inference limit

* fix bug

* reset install_cli

* make package buildable -- but llama.cpp features just do not work at this point because metal is not preprocessed

* add proper availability checks

* change macOS minimum version

* fix several problems

* code cleanup

* enable ubuntu build

* fix build error

* fix ubuntu build

* fix borrowing

* update install_cli.sh

---------

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

.gitignore

@@ -17,3 +17,4 @@ Package.resolved
 *.pyc
 .docc-build
 .vscode
+*.gguf