So, all other things being equal, you think that hitting 75% of the time for 4.5 damage is not much less impressive than hitting 80% of the time for 7.5 damage? It's nearly double the damage starting out is what it is, and though the relative size of the advantage might decrease, it's still there and the one with 18 strength as a fighter also gains 10% more XP than his less advantaged peer, and can in general gain more immediate advantage by investing wishes in wishes for higher strength, or by the use of manuals of gainful exercise, than could a character with 9-15 who has nothing. If built as a Cavalier, your training special ability gets massive and immediate return on investment compared to a Cavalier of minimum strength.
Not at all. Front line characters without Con bonuses simply cannot survive for long in 1e/2e AD&D. Generally speaking, 90 hit points is considered a good safe amount, though more is achievable and desirable. Among other things if you don't have that, you'll sooner or later run into a dragon or spellcaster where even on a successful save, you'll still die from half damage. And that's assuming your DM isn't of the RB school and isn't trying too hard - see my DM's guide to challenging high level AD&D characters. Meanwhile, if you play a 19 Con Dwarf fighter, you'll have about 77 hit points at just 7th level - something a fighter with 14 Con can't expect to achieve until about 17th level (which he will never survive to reach anyway).
First of all, if you have a high Dexterity, chances are you are building a specialized archer in order to get the double damage within 30' bonus and to be able to decimate just about anything before it can close with you. It's possible you are going to multiclass or dual class into thief, but the straight archer build (particularly as a ranger because you aren't forgoing the XP bonus and you have advantageous surprise mechanics) is very viable. It's not just the extra AC. Other than the attack adjustment, one of the biggest advantages that high dexterity gives you is it lets you dominate the surprise mechanics as an archer, peppering enemies with massive amounts of arrows before they can even react - and then hitting them again before they can react even if you lose the initiative. High dexterity archers are bad news, and if you haven't seen one pump out 60-100 damage during a surprise then you need to go back and reread the surprise rules.
But beyond that, you aren't really thinking through how AC functionally works in D&D. If something needs a 19 to hit you, then if you can squeeze one more point of AC into your build, you take half damage because needing a 20 to hit you means they hit you half as often. If something needs an 18 to hit you, then if you have 2 more points of AC, then you think only 1/3rd damage. And this is the secret weapon of PC's in 1e/2e AD&D - because most monsters don't have bonuses to hit, and because monster THAC0 is effectively capped, it's not that hard to get negative AC and then, just about nothing can actually smack you. The differences between an AC of say 2 and getting hit 1/2 the time, and -2 and getting hit a 1/4 of the time, and an AC of -6 and getting hit 1/20th of the time is enormous. As you close in on the foe needing that 20, the advantage gets greater and greater in a non-linear fashion.
The advantages of an 18 Wisdom over say a 13 Wisdom for a cleric are likewise enormous - higher effective cap on your level, much more bonus spells, faster leveling owing to bonus XP, and on top of that a +4 saving throw versus mental attacks. The same is true for 18 Intelligence versus say 13 Intelligence of a Magic-User. Intelligence Table II is harsh on those which it finds wanting. If you are used to 3e and the rules being fair, AD&D is going to come as a shock.
The result of all of this is again, if you didn't have a viable stat array, you rerolled until you got one. If your DM frowned on that, you took all the risks for the party until you did get a character worth keeping - or you managed to wheedle the DM into allowing a chargen system that reliably produced viable stat arrays. Or you cheated.