Redundant switches for self-assembly storage?

Question

Computer networks

At one end are two boxes full of disks: zfs for backups and non-critical virtual machines for internal needs (proxmox); in them on 4 normal network gigabit. The boxes already exist, waiting for ECC in purchases.
At the other end there is an adult piece of iron, but I don’t know yet which one (which one will give 2-4 free ports).
What and how to put between them for fault tolerance and throughput?
I understand so far this way: two cables from each box into two switches, of which 1-2 cables to a higher-level piece of iron.
What the budget allows: D-LINK DGS-1100-08 (16), HPE 1820-8G, ZYXEL GS1200-8HP, HPE OfficeConnect 1920S, D-LINK DGS-1210-10/ME.
In my head there is a light (heavy) mess of lacp, bonding, vlan, stp, trunks, etc.
Questions: do I see the diagram correctly; will it work as I want; what switch will be enough for these purposes. The question of how to set it up will not particularly stand after the answers to these, I’ll google it. The level, of course, enikey.

Answer 1

Why do you need switches in such a scheme and what kind of fault tolerance do you expect?
If you are allocated 1 port, install a switch, but there is no fault tolerance either.
2 ports - each server to its own port.
4 ports - same as with 2, only plus bonding.
Servers are additionally switched between each other if necessary.
The scheme with switches would make sense if there were 3+ servers. Reserved via rstp.

Answer 2

If you need stability, then why are you introducing an additional point of failure in the form of an intermediate switch ??

Answer 3

A pair of managed switches is enough.

Answer 4

In my head there is a light (heavy) mess of lacp, bonding, vlan, stp, trunks, etc.

lacp, bonding are essentially the same - aggregation, only everyone wants to stand out and comes up with their own unique name.
vlan, stp, trunks - troubles from the link level. Best dealt with in SDSM .
When choosing switches, you should pay attention to the size of the cache memory, the quality of the hardware and the stability of the firmware. In this regard, HPE is beyond competition.

What and how to put between them for fault tolerance and throughput?

And here you have to choose one thing (there is such a CAP theorem ). If you want to get both pigeons, then you fall under the requirement 3I (redundancy, redundancy and again redundancy).