I fully agree.
And I truly believe that the world would be a significantly better place if software developers worked for stability and fixing issues, instead of entertaining themselves with "cool" (to their understanding) features or migrations/refactors to new frameworks or languages (and yes, I know these are, sometimes, truly needed). We are way to good at justifying wasted effort and plain "play".
For offline software, is maintenance revenue enough to maintain a support team? What happens when most bugs have been fixed and customers stop purchasing maintenance? Is the business prepared to market old versions in parallel with new versions?
The meager support provided with software is an intentional tradeoff. I'm sure it isn't the right choice all the time, and as you say, there probably is demand for a different set of tradeoffs, but for example, I have no desire to pay Microsoft up front for 20 years of support, I'd rather get shiny new things more often than that.
Building owners (not builders) choose when to tear down buildings.
Software builders often choose to tear down software against the will of their customers.
Customers buy $100 shrink wrapped Windows (or $$ enterprise support for the same thing) instead of spending far more money on software that they would have more control over. So sure, the software companies are only living up to the smaller promise, but the customer didn't bother getting (and paying for) a better one.
But we don't yet know whether the market would pay for stable app versions that are encapsulated to remain compatible with new hardware and operating systems. The presence of client hypervisors on Windows 10 and OS X will enable new experiments. Perhaps a company can devise a marketing message, business model and toolchain which supports stability and forward-compatibility.
Unikernels could extend the shelf-life of applications, reducing the need to comply with OS pressure for upgrades. Mature and battle-tested applications need security upgrades and the occasional minor bug fix.
Imagine the societal chaos if buildings rerranged themselves without warning or just disappeared ("end-of-support"). We are missing an analytic framework to calculate the economic benefits of stability and costs of pointless change.
Windows XP has proven that customers will pay for not changing what works perfectly well. With virtualization to provide drivers for new hardware, mature OS/apps can live for decades.