Deal with LAVA callback tokens separately

[gctucker]

At the moment, the LAVA callback tokens are API tokens which are then used directly to send the data to the API. However, they are JWT tokens which may not always fit in the LAVA token database used for callbacks. In fact this seems to depend on the length of the user name... (kernelci-pipeline had a JWT token that didn't fit in the LAVA database. Now we have the kci user and the JWT is shorter). This is to avoid having a special kind of token for the LAVA callback handler.

One option would be to change how LAVA callback tokens are being handled as it's a bit of a hack, relying on the Django tokens and using the description to look up the correct one to use. There could be a separate database table for secrets or just callback tokens instead with an arbitrary token length (e.g. 64k).