CANParser: always parse all messages

Fix for problem 1 from https://github.com/commaai/opendbc/pull/2547

---

I think the fix for this is to speed up the parsing such that we can always comfortably parse all messages and signals, then maintain a list of actually read messages to use for the `canValid` flag.

---

This should also let us ship https://github.com/commaai/opendbc/pull/2549.

Start by reading problem 1 in PR #2547 and the proposed relationship with PR #2549; the issue does not identify files or tests. Done means CANParser can parse all messages and signals comfortably while canValid is based on messages actually read.

Written by the indexing model from the issue text.