add more checks for impossible constraints

[arvidn]

If a single spend bundle has an ASSERT_BEFORE_HEIGHT_ABSOLUTE that's lower than or equal to another ASSERT_HEIGHT_ABSOLUTE, the spend bundle will never be valid.

This patch checks for cases like this. Also for _SECONDS_ABSOLUTE.

The _RELATIVE checks can only be made within the same Spend, since we don't actually know any absolute timestamps or confirmation heights for the coins.

This is to catch a few obvious invalid cases early.

This still leaves non-obvious cases as successful (and must fail at a later state in the mempool pipeline).

For example, two different coins can have ASSERT_HEIGHT_RELATIVE and ASSERT_BEFORE_HEIGHT_RELATIVE that turn out to conflict, but we'll only know that once we look those coins up from the coin DB and know their confirmation heights.

Another, more subtle, case of invalid spend bundles is where an ASSERT_SECONDS_ABSOLUTE conflicts with an ASSERT_BEFORE_HEIGHT_ABSOLUTE. In chia_rs we don't have any mapping between block height and (predicted) timestamps. Clearly we couldn't reject spend bundles based on predictions anyway, but there may be cases where a SECONDS timestamp is known to refer to an old, existing, block. We won't catch those cases here either.

[richardkiss]

Seems reasonable. The repeated code feels a bit wrong, but I suppose in this case it's okay.

Alternative is we could just check every time through the loop, or just once at the end of conditions processing. I don't see a case where exiting early buys us much.

[arvidn]

Alternative is we could just check every time through the loop, or just once at the end of conditions processing. I don't see a case where exiting early buys us much.

Yeah, that was my first version, but it seemed hard to justify checking every time