Use defer_build with type adapters, not custom _LazyTypeAdapter

[sydney-runkle]

Fix #248

While using _LazyTypeAdapter is slightly faster, I think both operations are fast enough that we should use the recommended defer_build pattern here:

from pydantic import TypeAdapter, ConfigDict
from typing_extensions import TypedDict
from typing import TYPE_CHECKING
import time

if TYPE_CHECKING:
    LazyTypeAdapter = TypeAdapter
else:

    class LazyTypeAdapter:
        __slots__ = '_args', '_kwargs', '_type_adapter'

        def __init__(self, *args, **kwargs):
            self._args = args
            self._kwargs = kwargs
            self._type_adapter = None

        def __getattr__(self, item):
            if self._type_adapter is None:
                self._type_adapter = TypeAdapter(*self._args, **self._kwargs)
            return getattr(self._type_adapter, item)


class MyTD(TypedDict):
    a: int
    b: str
    c: bool


start1 = time.time()
ta1 = LazyTypeAdapter(list[MyTD])
end1 = time.time()
print(end1 - start1)
# > 1.1920928955078125e-06

start2 = time.time()
ta2 = TypeAdapter(list[MyTD], config=ConfigDict(defer_build=True))
end2 = time.time()
print(end2 - start2)
# > 1.6927719116210938e-05
# > Note: without defer_build, 0.14297986030578613

[cloudflare-workers-and-pages]

Deploying pydantic-ai with    Cloudflare Pages

Latest commit:	2c50868
Status:	✅  Deploy successful!
Preview URL:	https://3277ebed.pydantic-ai.pages.dev
Branch Preview URL:	https://defer-build-ta.pydantic-ai.pages.dev

View logs

[samuelcolvin]

otherwise LGTM.