GitLab

This makes it easier for OONI to receive our data.

Instead of passing our authentication token in the "auth_token" GET
parameter, we are now using the "Authorization" HTTP header as defined
in RFC 6750.

The advantage of this is that we don't have to worry about the URL
encoding of our Base64-encoded token.  Besides, middleboxes frequently
like to log request parameters (but not necessarily HTTP headers) and we
don't want our authentication token to be part of these logs.

Thanks to Arturo Filastò for suggesting this!

The article
<https://stackoverflow.blog/2020/03/02/best-practices-for-rest-api-design/>
convinced me that it's better to expose our bridge fetch API over a GET
rather than POST URL, and change the endpoint from "fetch" to "bridges".

For some reason, the following row ended up in BridgeDB's SQLite DB:

  "339948" "REDACTED" "None" "NULL" "moat" "2020-04-28 23:00" "2020-05-15 21:30"

"None" and "NULL" should be the bridge's IP address and OR port,
respectively.

This commit adds a clause to our SQLite statement that filters out rows
whose or_port is NULL.

We should log an error but there's no reason to terminate the service if
there's a transient error in reading bridges.

This makes it easy for us to monitor a wolpertinger instance.

This shortens the API URL from

  https://bridges.torproject.org/wolpertinger/api/fetch

to

  https://bridges.torproject.org/wolpertinger/fetch

If the given JSON structure cannot be decoded, the server would respond
with "EOF".  This patch replaces that string with a more helpful error
message.

Having our parsing functions expect a file handle rather than a file
name simplifies testing.