cabal-install should probably do a quick sanity check on downloaded tarballs

I sometimes develop in an environment where downloaded file contents often get silently replaced by error text in XML -- don't ask why.

In this situation, `cabal` will
- Download the corrupt `tar.gz`,
- Not create/download (?) a `tar.gz.etag`.

This is understandable so far.

Then `cabal` will merrily try to install (I assume) this corrupt file, and fail, but without producing a `log` file:

```
Failed to install zlib-0.6.1.1
Build log ( /home/thejanet/.cabal/logs/zlib-0.6.1.1.log ):
cabal: /home/thejanet/.cabal/logs/zlib-0.6.1.1.log: does not exist
```

It seems to me that `cabal` somehow _knows_ that the file was not downloaded properly, thus the lack of an `etag` and `log` file, so it would be nice if:
- `cabal` would give a better, more precise, error message that the file is corrupt, and
- It would try to re-download the file the next time I attempt to `cabal install` or `cabal fetch` it.

My main issue really, apart from the lack of a good error message, is the fact that the user needs to go and manually delete the corrupt download to prompt `cabal` to try again.

I appreciate that if we can't fetch the file then we probably won't be able to fetch its hash, but can we at least use a simple heuristic which at least tries to expand the file to make sure it is a valid `tar.gz`?

Reproduce the report by giving cabal-install a corrupt downloaded tar.gz and trying cabal install or cabal fetch. Trace the download, archive validation, cache, and retry handling; done means a precise corruption error is reported and the next attempt retries the download without manual deletion.

Written by the indexing model from the issue text.