Hello Giovanni
These are excellent questions, and I’ll try to address each one as well as possible.
Ideally, the access and distribution layers should not be in the same location. The access layer’s role is to be as close as possible to the end devices so that cable lengths can be limited to under 90m (which allows for 5m patch cords on each end) or much less if possible. The distribution layer should not be that close, but should be able to consolidate multiple access layer switches in various locations. If you have enough endpoints in a single classroom, then putting a rack in each room is not a bad idea (as long as you have redundant power, and you can successfully restrict access to the equipment). You can then use fibre to link multiple classrooms to a single distribution switch location without the limitation of cable distances.
Ideally, you should have an access layer. However, when money is involved it is often an incentive to compromise. in this case, it’s probably not worth the cost of an access switch with redundant power, and a rack and so on (which in some cases can be in excess of €1000), just for two computers. Connecting directly to the distribution switch is probably better, but is still an exception to the rule. Just make sure to document this clearly in both paper documentation as well as in the description of the ports in the devices.
The spine and leaf architecture is typically used more for datacenters rather than enterprise networks. If you are designing the datacenter as well, then you can examine this as an option, but for the campus network, you should stick to a more traditional 3 tier network.
Network redundancy is not typically designed into the access layer of a network. In order to achieve network redundancy for each workstation, you literally need to have two network cards on each computer, and connect each one to a port on a different network switch. This drastically increases the cost of structural cabling and the number of access switches you (double the number of ports and cables.).
However, redundancy can be achieved on the switches themselves by employing redundant power supplies and stackwise implementations that support sending power over the stackwise cable. This way even if a component of the access switch fails, the switch itself will not fail. This is a more financially viable solution, to beef up the access port switches, rather than provide redundancy all the way to the workstation.
I hope this has been helpful!
Laz