feat(host_fn): infer types from annotations

[chrisdickinson]

Infer host function Wasm types from type annotations if given. When inferred, parameters and return values are automatically marshaled and unmarshaled and "current plugin" access is hidden. Whenever a host function is declared, add it to a thread-local registry of host functions. Default Plugin creation will use the global list of plugins.

In the weeds note: because this means we have to hang on to functions at specific addresses, we always end up using user_data now: in particular we store the offset of the target host function in the user_data to determine which implementation to call. I added this because the FFI call would fail if the target python host function wasn't available at module-level; this would turn into a stack overflow (it appeared that the runtime tried to re-invoke a host function infinitely if it couldn't find it.)

---

Example of inferred decorator + global host func registry:

import typings
import extism
@extism.host_fn()
def incr_json(input: typings.Annotated[dict, extism.Json]) -> str
    return input.get("hello", "world")

with extism.Plugin(some_wasm_file) as plugin:
    plugin.call("incr_json", json.dumps({"hello", "earth"}))

[zshipko]

This is really nice!

The one thing I'm unclear about is if it's possible to a user to write their own encoding format (like Json or Pickle)? If not, maybe we could provide an abstract class users could implement - which also could be a big enough chunk of work for a separate PR. But for now it seems fine as-is since they could always use a string/bytes argument and decode it themselves.

[zshipko]

This is much more clear!

[chrisdickinson]

@zshipko thanks – and great idea! I've added an extism.Codec class for use as a type annotation in the latest commit.