CTxMemPool::GetMinFee should not return CFeeRate(0) #11455

The ::minRelayTxFee should be taken into account

This patch comes from #11410 (review) to include minimal code changes when fixing a bug.

Relates to #11410 and #6941

Please advise if I should set [fee] in commit message or something similar (thx @fanquake for label :) )

I ran fundrawtransaction many times, didn't get into problems. See #11410 (comment)

Next PR will cover ::minRelayTxFee in https://github.com/bitcoin/bitcoin/blob/e93fff1463ae906fc986bf98c3b118c82f171546/src/txmempool.cpp#L1002

~I don't believe this is safe to do, because a 0 fee transaction may be in the mempool as a result of a disconnected block.~

actually, with how GetMinFee is used, it's probably fine. Sorry for the noise.

Indeed, if there were an invariant in mempool that everything paid at least GetMinFee we'd already violate this (temporarily) in cases where mempool has been limited and we're mid-reorg, and more often when there are dependencies which result in something having a higher limit-enforced fee.

Still, I'd kinda prefer the original other option where the minRelayTxFee is used directly in getmempoolinfo instead of GetMinFee, in part because it breaks the wallet fee estimation "reason" value (see wallet/fees.cpp:69, which can no longer ever be hit in the fee estimation case as MEMPOOL_MIN will always be set, not a huge deal, but nice to not change that IMO)

I was about to comment that this should be maxed with incrementalRelayFee, as mentioned in #11410 (comment).

But then I don't see why we should keep around both (incrementalrelayfee and minrelaytxfee), as I don't see the difference after removal of pay-by-priority.

I think my suggestion to create a separate commit for this fix was wrong, as it is not really a bug, but more a inconsistency. I feel we should make the minrelaytxfee command line option an alias of incrementalrelayfee.

Might be better to close this and focus on the whole picture (#11410) again.