Bots should only receive messages targeted at them

*Note*: Some bots legitimately need access to *all* messages, and we should still permit those cases.

The general case for a bot is that it responds to commands by reading every message and doing something if it thinks the message matches a predetermined set. This is obviously not great as less trusted bots *could* become loggers too, though there is certainly an argument about only adding trusted bots to rooms.

This is more of an issue with encrypted rooms, where the bot has no reason to see messages not directed at it and can even be excluded more easily.

A challenge is doing this in a way where the bot can identify itself from a regular user (given bots are just user accounts), and not exposing metadata to the bot (ie: in an encrypted room, the bot doesn't need to know that *a* message was sent either).

Related:
* https://github.com/matrix-org/matrix-spec/issues/239
* https://github.com/matrix-org/matrix-spec/issues/56

No file, test, or entry point is named. Start by reading the issue alongside related issues #239 and #56, then determine how bots can receive only targeted messages while preserving legitimate access to all messages and avoiding metadata leakage in encrypted rooms. Done means a settled protocol design and documented specification change.

Written by the indexing model from the issue text.