Make timeout/retries configurable for HTTP transports

`cabal --help` shows support for a few transport methods:

```
--http-transport=HttpTransport Set a transport for http(s) requests.
Accepts 'curl', 'wget', 'powershell', and
'plain-http'. (default: 'curl')
```

Right now the timeout/retry configuration seems to be adhoc, e.g. `wget` gets passed hardwired `--tries`/`--timeout` values.

```haskell
wgetTransport :: ConfiguredProgram -> HttpTransport
wgetTransport prog =
HttpTransport gethttp posthttp posthttpfile puthttpfile True False
where
gethttp verbosity uri etag destPath reqHeaders = do
resp <- runWGet verbosity uri args
(code, _err, etag') <- parseResponse uri resp
return (code, etag')
where
args = [ "--output-document=" ++ destPath
, "--user-agent=" ++ userAgent
, "--tries=5"
, "--timeout=15"
, "--server-response" ]
++ concat
[ ["--header", "If-None-Match: " ++ t]
| t <- maybeToList etag ]
++ [ "--header=" ++ show name ++ ": " ++ value
| Header name value <- reqHeaders ]
```

Whereas e.g. `curl` is left at its default settings. For the mirror-fallback logic (e.g. when shall the code consider hackage.h.o to be down and cycle to the next available mirror?) we may want to have more control over these settings in order to allow the user to fine-tune/experiment to his environment (as well as provide us empirical data to inform the selection of good defaults).

/cc @gbaz