[java] loss of precision with `number` fields

We should set `useBigDecimals = true` in the jsonSchema2Pojo gradle config.

Presently, the protocol defines fields which are oddly of type `number`:
- `emitted_at` in TRACE and CONTROL messages
- `record_count`

The `emitted_at` field is always a millisecond epoch timestamp. In Java, the jackson library will always serialize `double` and `float` values using scientific notation, which can cause loss of precision in the least significant digits, yet these are precisely those which are interesting to the user here.

The `record_count` was deliberately picked to be a `double` though the reason isn't immediately obvious, a `long` should be quite sufficient.

In any case, the values are always integers and not terribly huge ones either, they should be perfectly safe to represent using `BigDecimal`.

This change is not backward-compatible, but perhaps that's not that big of a deal? I don't know.