Added fallback behavior to custom v1 resolvers

[isaiahvita]

When a custom legacy endpoint resolver (pre-EP20) was set via one of the config setter helper functions, it was wrapped with a fallback endpoint resolver
https://aws.github.io/aws-sdk-go-v2/docs/configuring-sdk/endpoints/

This alters that previous fallback behavior to fallback to the endpoint resolver v2

[lucix-aws]

question: Does downstream code handle this properly? i.e. if the v1 middleware gets a zero-value endpoint, it does nothing and lets v2 pick up the rest?

[isaiahvita]

it didnt. pushed new commits with a fix. and tested via debugger with sample program and unit tests. heres my sample program:

package main

import (
	"context"
	"log"
	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/config"
	"github.com/aws/aws-sdk-go-v2/service/s3"
)

func main() {
	customResolver := aws.EndpointResolverWithOptionsFunc(func(service, region string, options ...interface{}) (aws.Endpoint, error) {
		return aws.Endpoint{}, &aws.EndpointNotFoundError{}
	})
	
	cfg, err := config.LoadDefaultConfig(context.TODO(), config.WithEndpointResolverWithOptions(customResolver), config.WithRegion("us-west-2"))

	client := s3.NewFromConfig(cfg)

	_, err = client.GetObject(context.TODO(), &s3.GetObjectInput{
		Bucket: aws.String("test-bucket"),
		Key: aws.String("test-key"),
	})
	if err != nil {
		log.Fatal(err)
	}

}

[aajtodd]

Would be nice to see a unit test added to cover this behavior if possible.