Enable accessing Variables from the top level of the DAG files

[ashb]

Fixes #45449

Since I want to maintain the property of being able to run a DAG processor
without the Execution API server running, and since the Dag Processor Manager
already has a database connection I have chosen to run the FastAPI execution
server in process.

To achive this I make use of two features:

• The first is the abilty to provide an httpx.Client with a Transport object
  that has an WSGI appliction to not send real requests, but to instead call
  the WSGI app directly to service the request
• The second is a2wsgi. Since we are making a call from with in a synchronus
  context we have to give httpx a WSGI (if we were making Async requests we
  could give httpx an ASGI app directly), and FastAPI at it's outer layers is
  an async framework (even if it supports running sync routes) we need to
  somehow wrap the async call to return a sync result. a2wsgi does this for us
  by using a async loop off the main thread.

I tested this with a simple DAG file initially:

import time
import sys
from airflow.decorators import dag, task
from airflow.sdk import Variable

from airflow.utils.session import create_session

if Variable.get("hi", default=None):
    raise RuntimeError("Var hi was defined")

@dag(schedule=None)
def hi():
    @task()
    def hello():
        print("hello")
        time.sleep(3)
        print("goodbye")
        print("err mesg", file=sys.stderr)

    hello()

hi()

If the variable is defined it results in an import error. If the variable is
not defined you get the DAG defined.

---

^ Add meaningful description above
Read the Pull Request Guidelines for more information.
In case of fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
In case of a new dependency, check compliance with the ASF 3rd Party License Policy.
In case of backwards incompatible changes please leave a note in a newsfragment file, named {pr_number}.significant.rst or {issue_number}.significant.rst, in newsfragments.