The 'demonstrably untrue' part was the claim that 'point-buy lets players play the concept they want'.
Which is, itself, demonstrably untrue. ;P
The rest is just pointing out the near infinite concepts that point-buy absolutely disallows, and that those same concepts are still available when rolling; even if you have to wait for the right random roll, a small chance is still higher than zero chance.
In the sense that the Lotto will get you millions.
Yeah, random generation lets you hope, but it still necessarily gives you exactly one array to arrange from the myriad possibilities. You can give it credit for the myriad, but you only get the one. The degrees of freedom you have to build the character you want with random-and-arrange is identical to that of standard array - and less than that of point buy. That's just fact.
The flip side of that, which you're kinda alluding to, is Maxperson's attitude, not random lets you play what you want, but that you want to play what random gives you.
Which is fine, too.
If I generate a set of ability scores, by whatever method, then I try to create a concept that makes sense for those scores
Yep, random is excellent for sparking a concept idea, that way.
or try and assign scores that match my pre-existing concept.
Random isn't so great for that, on par with standard array: you get to shuffle scores around.
If I wanted to stat up The Avengers, Thor's Strength will be much higher than Captain America's, even though 'strong' is a feature of both concepts.
Yep, and if you tried to do that in D&D, you'd be out of luck, because Thor's strength is right off the D&D scale. OTOH, you'd have no problem statting out both Thor & Cap in Champions! which uses (surprise) point-buy, not just for stats, but for everything - Thor's super-STR, magic hammer, & invulnerability, Cap's human-paragon abilities, skills & shield.
The numbers do matter! The numbers are how the game quantifies a major part of our concepts, because how strong, dexterous, tough, smart, wise and charismatic are part of every concept, even if a particular ability is 'unremarkable' in a particular concept.
Numbers matter, of course, but they're relative. In a super-hero game, Thor & Cap (and Black Widow and Quicksilver) are all do-able. In D&D, they're not. In some D&D campaigns, a concept with very high stats across the board might be OK, or even expected, in others something more moderate might be preferred - that informs the method the DM chooses, but each method can give higher or lower stats, in general, it's just a matter of calibrating it to the campaign, using a different array than the standard, rolling more/fewer dice, or giving more/fewer points. Random carries the risk, however minute, of delivering a campaign-inappropriate array - that doesn't mean it covers more concepts, it means it fails some of the time (the character will need to be rejected by the DM and re-rolled).
In every iteration of D&D, that context is and has always been the 3d6 bell curve. That context has never changed.
It's just not the only context. 3 STR/18 DEX is 'weak but agile,' sure, relative to the curve, extremely so. So is STR 8/DEX 14. So is STR 10/DEX 12 in a party where the next-lowest STR is 16 and the next highest DEX 8. STR 19/DEX 12 might be 'weak but agile' for a storm giant, STR 8/DEX 15 might be strong but clutzy for a quickling.
The most important context for a PC is probably his party (and, thus, the method used to generate them), then the enemies he meets, then the NPCs, then the background world, then, finally, the hypothetical curve that started it all.