Not unlike living organisms, electronics have some ability to continuously self-repair. For example, dynamic random access memory stores energy in billions of tiny capacitors. These charges decay, and typically a self-refresh process continually sweeps through the memory reading and rewriting the rows of data before the charges decay too much.
Error correction codes are also used in some applications. Many can detect two-bit errors and correct one in a given chuck of data. If this technique is used, and a bad bit is detected and corrected before a second bit in the same chunk of data is corrupted, then the bad bits can be repaired.
The technique can even be used by striping the chunks of data across banks of parts, such as in the Redundant Array of Independent Disks (RAID) arrays used in the servers in most data centers. In this case, if a bad hard drive can be detected and swapped out quickly, before a second drive fails, then all of the data can be recovered from the remaining working drives.
Built-in-self-tests (automated versions of "running a diagnostic" in Star Trek) can also detect failures and bad silicon can be taken out of service by the operating system. This can be a useful strategy if a bad core is detected in a multi-core processor, for example. Or, bad sectors of a storage device can be detected and flagged as "do not use".
In some cases, such as what happened in April 2024 with Voyager 1, clever engineers can figure out what went wrong and figure out how to work around the problem. In the case of Voyager 1, it took them 5 months, but considering that the craft was 24 billion kilometers away and was designed in the 1970s, that's pretty impressive. Fortunately, errors that require this level of intervention are less frequent.
There are lots of techniques along these lines that can help to protect electronics, but most of them can be overwhelmed if the errors occur too frequently or if the self-repair techniques don't repair quickly enough.
Total Ionizing Dose is a good proxy for how frequently errors are likely to occur and could help an electronics engineer figure out if their hardware's various self-repair techniques would be overwhelmed.
As for why chips used in space are radiation-hardened, the reason is simply that none of the self-repair technologies developed for the commercial market are particularly foolproof, since it is relatively easy to swap out parts and back up data down here on Earth. Space is a different environment with different rules. The term "radiation hardened" really just means creating electronics that are better adapted to these different rules.
In practice, a company can design a chip that makes liberal use of all of the self-repair technologies we know of and even invent some new ones. Engineers will then need to do lots of testing in a synthetic radiation environment to verify that it is robust, diagnose why the chip fails despite their best efforts to make it bulletproof and try again. Repeating this process many times will lead to a radiation-hardened chip. But, it will be an expensive and time-consuming project. There will be very few customers over which to amortize the Research, Development, Testing, and Evaluation (RDT&E) costs.
So, a radiation-hardened computer can end up being crazy expensive (for example, $338,000 for a RAD750 which was used in JWST) and probably will not be very state-of-the-art when it comes to performance and modern features (other than self-repair features). But, it may well be worth it to reduce the overall risk of mission failure.
I think that all three of the hypothetical reasons you listed are correct.
Radiation levels can be correlated with chip error resilience (as explained above).
Extra mass for shielding can be incredibly expensive (around 1.2 M per kg for one-way trips to the Moon and Mars). But, due to cosmic ray interactions with shielding materials, shielding can cause a single particle to generate a whole shower of particle interactions whereas before there would be only one interaction per cosmic ray. So, the benefit of just a limited amount of shielding is not clear-cut.
A requirement such as "Shall survive worst-case space weather" is very likely to make it onto a product requirements document, which means the selected part will almost certainly be overdesigned for typical space weather conditions.