feat #6: Track number of requests per user agent

[MarcusRiemer]

This is a first example on how working with one large table could look like when tracking user agents. Notable things from the top of my head:

• I somewhat attempted to make the new columns optional. If someone doesn't run the migration but updates to this version, everything should work as it did before. But I wouldn't know how this could be properly tested.
• Making user agent tracking optional was a little bit annoying for the Redis queuing that I missed initially, as I don't use it. I took the liberty to refactor the serialization and deserialization of such keys but didn't include any migration path because I have no idea how to test this properly. So users would need to drain the queue as part of the migration. I guess this could be salvaged by checking whether the key is in the old format if this feels important.
• Customization of extracted strings is hinted at, but not yet implemented. But this should be mainly a documentation issue, as the code is technically only missing a setter.
• I bumped the minimum Rails version to 6 in order to use SQL upserts. This should be more performant and more secure (for whatever security is needed to track single visits, but hey). This however has an important implication: As in SQL NULL is not equal to NULL the upsert will only work if all columns have empty strings as default. This is not yet reflected in the migration.
• I didn't touch the UI for the moment.

[MarcusRiemer]

The aggregation has to live in Ruby for the moment because I didn't want to hard code which exact values for browser, operating_system and device_type we want to allow. Technically this would be currently know from my implementation of user_agent_extractors.

But it didn't feel right to add / generate mildly complex SQL to aggregate exact values when not knowing whether this might just be good enough.

[alexisbernard]

Is there a limitation to use browser.name, browser.device.name and browser.platform.name ?

[MarcusRiemer]

No, this was just me wanting to make 100% sure I know all the possible values that could be written. Just as you mentioned (#6 (comment)): The potential for far too many rows is big.