At this moment, the total hashrate is 19500 Ph/s.
In order to perform a 51% attack, you'd need a mining farm producing ~10.000 Ph/s, that's 10.000.000 Th/s

A single antminer S9 hashes at 14 Th/s.

So, you'd need about 700.000 S9's to perform a 51 attack right now.. The diff would have to drop a whole lot before a 51% attack is feasible.

That being said, your concern isn't completely unfunded... The theory is right. IF, for some reason, the hashrate drops steadily to (for example) 0.001% of the current hashrate, the diff would follow, and at that point, 7 S9's would suffice for a 51% attack. I do think that if this ever happens, there would be something seriously wrong, and a 51% will probably the least of our problems.

In theory, you're right... In case of a 80% drop 150.000 S9's will be sufficient... However, buying, setting up, running and maintaining 150.000 S9's is a collosal (and expensive) task. I mean, even after setting everything up, the power draw of such a mega farm would be enormous...

The price of 150.000 second hand antminers would still be over 10 million USD (my guesstimation), you'd need thousands of racks, you'd need special serverrooms with cooling, tech guys, cabling,... Just to be able to perform a 51% attack. A 51% in itself is also limited regarding the amount of history that can be rewritten. Even if you controll 51% of the hashrate, i don't think it would be wise to try to re-mine the last 10 blocks...
My point is: i think it would be pretty obvious who is executing a 51% attack if this person hash to spend multiple millions and dozens of tech guys to setup his infrastructure. I think that at this point, most people *might* have chosen the path of an altcoin that doesn't use sha256d as it's POW algo.
If word gets out that somebody untrustworthy might be preparing for a 51% attack, i guess most people will switch, lowering the bitcoin price, lowering the FIAT equivalent amount of BTC the attacker can effectively steal, making the whole endaveour even less profitable.

As for your second question: i don't know what will happen after all coinbase rewards are claimed. The theory is that bitcoin will be so valuable that miners will earn enough from mining fees. In reality, the reward halves every ~4 years anyways, so i think that sooner or later there'll be an equilibrium anyways. In my opinion, it won't take another 3 halvings before the income from collecting fees outweighs the income of the block reward anyways. Combine this with innovations on the ASIC front, changes in the protocol, regulation or deregulation, banning, changing electricity prices and it becomes completely impossible to make a prediction about the future of mining.

So i guess there is no point in organizing such attack because if it happens, it will destroy all the value of the whole Bitcoin system.
indeed, that's more or less what i was trying to say

I understand the idea about millions of dollars for a equipment buyout (though i believe that for already repaid investment in equipment the selling price might be even less than 1000 dollars).
I'm sure it would be less than 1000 USD, but you'd have to buy an S9, a PSU, cables, shipping and handling and import tax. Let's say you could get a used S9 for $200, i think it's fair to say that that would be a bottom price, since a decent PSU is already +$100/piece.
200*150.000 = 30 million USD. My $10M was a gross underestimation

But what about electricity consumption?
1 S9 draws 1,4 Kw. 150.000 * 1.4 * 24 * 365 = 1800 MWu per year... Even if you get this electricity at 5 cents/Kwu, you'd still have to pay $91 million per year for your electricity.

Also you say about limitations regarding the amount of history that can be rewritten, what do you mean? As far as i know there are (or at least were) no such limitations. so all the history theoretically can be rewritten.
You still need the longest chain. if you want to rewrite the complete history, you have to catch up all blocks untill you have the longest chain, if you only own 51% of the hashrate, that would be a very hard thing to do...

Speaking about equilibrium between reward and fees - that's possible, but the bitcoin will may become so valuable, that the fee will be too high for costumers to continue use bitcoins as a payment system (its original purpose). Than the price will more and more depend on the necessity to return investment in mining equipment. it will go higher and will fall after all coinbase rewards are claimed. And so on and so further.
Can't really comment on this... It's so far in the future i don't know what will happen

Also another idea - how can we be sure that if this attach happens and one will be able to obtain some crypto gold, the rest will drop off it? it will still bitcoin, and its fame is spreading fast across the world, so maybe some investors will still try to hold it.
I don't know what will happen, but if a disaster like a very big 51% attack is happening, i doubt the markets will react favorable... My best guess would be that you see a lot of dumping and plumeting prices if this would ever happen.

EDIT: i just wanted to expand my answer on the longest chain...
For example, let's assume you had all those antminers and you wanted to re-mine all blocks, you'd have to adhere to the same set of rules regarding difficulty, targets, POW algo, block reward,... If you'd change any parameters, the blocks you mined would no longer be valid for the other nodes, and you would have created an altcoin. For the sake of argument, let's assume you get all other members of the network to remove the checkpoints from their wallet's sourcecode.

Now, let's start by completely remine all blocks.
The genesis block remains the same, but block 1 has a diff of 1... Your mining farm mines all 1439 blocks in a matter of seconds... After the 1440 first blocks, your difficulty jumps to 4... bummer... Your big farm still succeed in mining the next 1440 blocks in a matter of seconds but at that point, the diff becomes 16.
Every 1440 blocks, the difficulty goes up 400% (which is the maximum adjustment IIRC).
Now, it wouldn't take to long before the time between 2 blocks mined by the 51% attacker was also 10 minutes... Since the nodes are programmed in such a way to always pick the longest chain, it'll be impossible to achieve the longest chain if your "complete rewrite"-chain has an average time between blocks of 10 minutes, and the "old" chain also has an average time between blocks of 10 minutes but the "old" chain has a head start of over 500.000 blocks.