Harden publishing

[marta-lokhova]

This PR contains two changes:

• Properly fsync HAS files that act as a "publish queue". Note that the checkpoints themselves are using XDR streams with fsync turned on already, so this change impacts HAS files only.
• Gracefully handle situations where nodes enable publish mid-checkpoint. Core will wait until the next checkpoint to begin publishing. Without this change, nodes would need to rebuild state before enabling publishing (although it's unclear how often partial archives occur in practice).

Additionally, I added a few more thing to aid debugging, such as printing checkpoint files with dump-xdr and an offline command to dump all checkpoints scheduled to publish.

[SirTyson]

LGTM after the MODE_DISABLES_FSYNC change.