Refactor trigger decorator and deploy time evaluation logic

[talsperre]

Refactored @trigger logic and simplified the deploy time evaluation logic of @trigger decorator. The current implementation is buggy and fails when running trigger4.py in this gist here.

For testing, refer to all the flows in this gist here. We can add few more tests as well if needed.

Change in contract:

Previously, we allowed deploy time functions (not documented anywhere) at various levels in the @trigger decorator. So something like this was possible:

@trigger(
    events=[
        {
            "name": myfunc1,
            "parameters": [myfunc2, myfunc3]
        },
)
class MyFlow(FlowSpec):

Now, we only allow specifying all of parameters as a deploy time function:

@trigger(
    events=[
        {
            "name": myfunc1,
            "parameters": myfunc2
        },
)
class MyFlow(FlowSpec):

These are also still allowed:

@trigger(events=myfunc)
class MyFlow(FlowSpec):

@trigger(events=[myfunc1, myfunc2])
class MyFlow(FlowSpec):

[nflx-mf-bot]

Netflix internal testing[4457218] @ 3b7ef8e

[nflx-mf-bot]

Testing for OSS PR [1052] @ commit 3b7ef8e had 11 FAILUREs.

[romain-intel]

A few minor comments. I haven't tested with the test flows yet but will look.

[romain-intel]

this error message does not seem to properly reflect the condition here. You are actually checking the the content of the tuple are also strings. That too, this is a list of strings or tuples.

[romain-intel]

This also seems like a wrong comment based on the error condition below.

[romain-intel]

return type seems too permissive. You are either returning a callable or a Dict[str, str]