Support for ia:// (without need of ffspec dependency)

[lfoppiano]

This PR implements the issue #16 (it ships additional changes that are supposed to go into ffspec).

The implementation does rely on the HTTP backend, and not on the S3 one. The reason is due to the limitation in the IA service that does not allow ranged queries.

The workaround (implemented) is to not use IA's S3 endpoint at all. ia://item/file is rewritten to https://archive.org/download/item/file and read as ordinary HTTP.

Requesting a resource via the s3 path, the service returns 200 OK with the full 1.6 GB body instead of 206:

Request	Answer
curl -r 0-99 to the data node directly, http	200, Content-Length 1689855303
Same over https	200, full length
Explicit Range: bytes=0-99 header through the front end with -L	307, then, when following the redirect 200 with full length

The IA service provides Accept-Ranges: bytes, but IA's own documentation states "HTTP 1.1 Range headers are ignored" (see: https://archive.org/developers/ias3.html#how-this-is-different-from-normal-s3)

I could not find the sources of the IA service, but looks like there are other incompatibilities with S3 (307 mismatch with the S3 convention). Result: using botocore will result in having thrown an exception of unbounded recursion.