Technically all that is needed for a hard fork is 51% of sustained hashing power, any number above that is completely arbitrary, but does change the social contract a bit IMHO... when devs choose 75% of the last 1k blocks to execute a countdown for deployment it is more democratic in nature where a majority can coerce or ostracize a minority. Selecting a higher threshold of 95% is much more anarchistic in nature where one needs near complete consensus before the consensus rules are changed in a hardfork.

What I find troubling is I keep hearing many misinformed users who are under the impression that the miners decide on consensus rules when in fact it is the nodes that vote on consensus rules.

The Vote is = The longest valid PoW chain, with an emphasis on valid and only nodes deciding what is and isn't valid. The miners can hash all they want on an invalid chain , but they won't be necessarily following the nodes or economic majority.

This matters because when you look at how the hardforks are rolled out they are not polling the nodes but miners for an rough and indirect means of determining node support. There very well could be a dangerous situation where 75% of miners decide to fork , but over 75% of nodes and the economic consensus decides not to which would be very dangerous.  This is why I recommend that all hardforks have at least 95% consensus of blocks within the last 1k to activate a countdown. Luckily Bitcoin classic still has time to change their minds from 75% to 95% as they have yet to release code.