This asserts the initialization of a Hydra head is also seen by the (not
yet existing) hydra-chain-observer executable

We might want to implement that strategy as an observeRawCollectTx
first and see whether that aligns with the "full" observation.

We plan to use the observeCommitTx also from the chain observer. To make
that happen we need to remove stateful information like the known
initials of a head.

However, those are (in a contrived way) currently our only way of
ensuring that this transaction is a proper Hydra Head transaction.

The FIXME describes alternatives how to achieve this via keeping the
seed in the datum or providing the full resolved inputs of a
transaction (including the committed UTxO).

This ensures we do not observe transactions which look like commit
transactions on their outputs, but do not spend from initial script.

This ensures that the commit tx is part of the Head protocol (because
the validators must have ran).

This avoids passing the "too big" type into the HydraClient

This package is released with all other hydra-* packages and shares a changelog

This makes it clear that the hydra-node still only "cares" about commit
transactions about the head it knows in the chain layer.