When I hear that kind of characterisation, that something is a toy because it only offers relatively modest capabilities versus something high end, I can't help but think the person making the characterisation either doesn't know how to design things to be efficient and needs all that real estate just to get something simple done (e.g. we all know people who take megabytes and lots of MIPs to do something that many of us can do with a few bytes and a few hundred CPU cycles AKA "python programmers" ) Or they are compensating for other feelings of inadequacy - cf "big car syndrome". Or it's like the boss who needs the biggest and fastest computer in the office just to read emails and surf the web, pure self aggrandisement. I'm not saying that is the case here, just that it's what that kind of posturing always makes me think.
Time to market is just as important as resources required, that's why Python is a de-facto standard in data science, ML/AI, capital markets and other areas where getting results fast is more important than optimizing resources required. I don't use that approach and do optimizations when warranted, but there are cases when you simply can't do certain things with those super low-end toy FPGAs. Since I do a lot of image and video processing, the dividing line for me is ability of FPGA to natively handle 1080p@60 video. None of those toy FPGA can do anything in this regard to any meaningful degree, as simply running all logic and hardIPs like BRAM and DSPs at 150 MHz is often a challenge for them.
They still have a niche as companion chips in larger designs (as I remember ice40 were/are used in iPhones to control notification LED), but as far as their usage as a main processing chip they simply lack the speed and resources for most modern tasks.
As far as FPGAs that have modestly sized and modestly priced breakout boards available (on the same scale as bluepill and blackpill boards for STM32 microprocessors) there are quite a range of boards for the Lattice iCE40 series of FPGAs. I won't recommend specifics, as the one I have personal experience of, the original upduino, is no longer available (it was all of $7 for a complete board with a 5k LUT FPGA!!!). There have been other boards produced under the 'upduino' name that are currently available. Typically they have 4k or 5k LUTs wihich is a useful middle ground, not so large as to be expensive or encourage wasteful inefficient implementation ( ), not so small that you'll be fighting to find enough LUTs to do more than implement relatively trivial logic.
You are behind the times when it comes to FPGAs. Artix-7 and Spartan-7 are low-end nowadays (Xilinx politely calls them "Cost-optimized" even though some of them still cost few hundred bucks
), with Kintex-7/Virtex-7 and Kintex Ultrascale being a mid-range and Ultrascale+ series is high-end. So tiny toy FPGA like Lattice UltraPlus are not "modestly sized", they are ultra low-end. It's just the fact that even midrange devices are out of price range for all but the richest hobbyists, makes those toy FPGA seem like anything more than what they are. Heck, Artix-7 was introduced 11 years ago! Do you remember what CPU was high-end at that time, and how it ranks now? It's no different with FPGA, just the pricing scale is VERY different. Just because I (and most others) can not afford high-end devices doesn't change the fact that low end is a low end.
An upside for those who care is that the iCE40 range has a completely open source toolchain available for it. I have no philosophical axe to grind on that front, but the availability of a toolchain that is all command line orientated rather than GUI oriented better suits my way of working, and may better suit other folks of a similar bent.
All vendor-provided toolchains I know of are command line based, and have integrating scripting language (Tcl) to aid in automation. I know this for sure of Vivado, but as far as I remember Quartus is very similar in this aspect.
--------------
As for what's good for beginners, I think something capable of displaying a fullHD video is a good place to start because it offers a very visual feedback, which aids in maintaining your interest in this topic. Blinking LEDs becomes old very quickly, so having something more visual is much better in my opinion. At least it's been the case personally for me.