Requires api_helper_indexes plugin to start node

[abitmore]

#1836 introduced a wlog in constructor of database_api_impl, this PR changes it to dlog because wlog will cause disk space issue on API nodes (see #1799 and #1798).

[pmconrad]

This is a warning on purpose. API nodes should enable the plugin because otherwise clients will receive incomplete information. Changing the log level to debug means the node owner won't notice his configuration problem.

[abitmore]

Perhaps we add an option so the admins who really want to disable the plugin can disable the logging?

[pmconrad]

Hm, it feels wrong to add an option for a single log message.
IMO node owners who want to run API nodes without that plugin can post-process the logfiles to remove the message if it annoys them.
I can't really imagine a use-case where that makes much sense though. An API node is either public and should enable the plugin, or it's private and sees little API traffic and therefore only few of these messages. (Or maybe it's a private power user with a lot of API traffic - but I'd expect a power user to be able to tweak logging the way he wants it.)

[abitmore]

I think it's better to crash the node on start if we want to force the user to enable that plugin, than crash the node due to disk space issue after it's running for a few days. Better bother the admins once than twice.

Then, it would be a good idea to add an option to avoid the crash.

[abitmore]

Force pushed. @pmconrad please review.

[pmconrad]

IMO this option is annoying for non-API node admins. Please move the check into application::reset_websocket... so the option is only required for nodes that actually provide API endpoints.

[abitmore]

application::reset_websocket() is buried too deep in the call flow, so I added the checking in main(), the outcome should be the same.

[pmconrad]

Looks good, thanks!