1. Microcontroller selection. This can be a daunting task. Best to just pick one with good dev kits and community support. You can tailor the part number if you need more IO for a specific project. The only real choice to make is if you want/need to work in the low-power embedded realm (eg Arduino or STM32), or in higher-power/userspace realm (eg raspberry Pi or Beaglebone/Linux).
2. Rules of thumb for the design. You mentioned some great ones. I would add - use 4 layers unless you need more (and you will know if you do) - Put signals on the outer layers so you can access them for troubleshooting, unless you have a reason to have ground planes on the outer layers (you'll know if you do) - Use the internal 2 layers for ground and power. The keyword "polyfill" will let you know how to do this. - Don't try and make PCBs at home. It's so fast and cheap to order them. And a home-made one, even if you make it with a router instead of the printer/acid method, will not have a solder mask. You can easily spend 3 days troubleshooting. Watch movies for 3 days while you wait for your delivery, instead. - If you have any doubts, use a breadboard for prototyping. If you have confidence on your 3rd board design, skip the breadboard. - Put a hole in each corner, with a healthy clearance from any circuit to a fastener. At some point you will want to mount the PCB somewhere. - Put some spare IO on a connector if you have space - Use a barrel connector or USB for input power if you can. If you need more than 5V, USB-C at another voltage is trivial to add these days (as this page shows). It's a matter of when, not if, you mess up your settings on your bench power supply. - Use 2.54mm pin headers where you can instead of weird connectors - Ask yourself: Can I buy something that will work instead? - Invest in 3 kits: surface mount resistors, surface mount capacitors, through-hole resistors. eg: https://www.sparkfun.com/products/10969 - Spend at least $100 on a soldering iron with temperature control
There are a whole bunch of improvements I'd suggest with this particular design and article though: (also, doing six respins for a board this simple is a bit painful, and I'd encourage people to submit their layout to someone for a design review before going through this many iterations - I've reviewed things for friends in exchange for a pint of beer!)
- this board has no ground plane, and no return paths near the data traces - this will be an absolute EMI nightmare and also signal integrity will suffer. At very least, dedicate the bottom layer to ground (and flood-fill it) - the return path for digital signals (and anything above a few KHz) is directly below the signal path - see https://incompliancemag.com/wp-content/uploads/2017/04/1705_....
- (the clock design, in particular, is completely against ST's recommendations - there's no ground plane and traces running directly underneath!)
- panelisation: as others have mentioned, even for large-scale production runs, the PCB manufacturers will do a better job of panellising themselves, and can stack your boards in amongst other boards on the same panel if you're not taking up a whole panel.
- unless you've got a good reason to do it, having a serial programming interface with a serial/USB chip isn't the easiest way of doing it - exposing SWDIO/SWDCK/SWO to a suitable connector and using an ST-Link in-circuit programmer is generally an easier and more space-efficient way! They seem to have gone that way in the last couple of revisions.
There are a bunch of other things I'd probably do differently, but this is all to say - if you do decide to take up this hobby, please do get your designs looked at by someone knowledgeable, you'll learn a lot that way!