Very good question.

Let's break it down:

> A bank give a Merry loan of 100,000$

Concretely, the bank records an asset of 100,000 (the loan) and a liability of 100,000$ (the amount recorded in Merry's deposit account).

Note that at this point, the liability is what appears on your internet banking statements as 'money' and those liabilities to you from banks are today what is traded in exchange for goods and services. In this sense the liability is 'money',  and it is why people consider this money to have been synthesized commercially *ex nihilo*. This is essentially correct. Banks are licensed to create this money from nothing.

> Merry buys a house with this money. 

Analyse it in detail: Merry finds a supplier, who implicitly trusts the bank, who probably has an account at another bank (though not necessarily), and agrees to accept payment in the form of a bank liability. Merry initiates a bank transfer from MerryBank to SupplierBank. MerryBank decreases its liability to Merry by 100,000, keeping only the asset, but adds a new liability to SupplierBank. SupplierBank records an asset of 100,000 (MerryBank's liability to SupplierBank) and a liability to Supplier of $100,000. 

> For a period of 1 year, Merry has paid back 10,000 to the bank. 

Yes, in 99% of cases in the form of bank transfers from other newly created debt from other banks. These become assets in the books of MerryBank (liabilities from SomeOtherBank) and the loan asset is reduced. So while MerryBank gains a 10k asset from SomeOtherBank it actually cancels 10k of loan asset, so its position has not changed - only its risk position.

> Afterward Merry's finance condition worsen and she no longer can pay
> her debt. So in theory, if all the bank borrowers end up in the same
> situation as Merry did, the bank is going to bankrupt.

No. The bank still has an asset of 90,000$ and will use debt collectors to take Merry's property. The bank will then cancel the 90,000 loan asset and record a property asset at market value.

> But if the bank never had the loan money in the first place,

Yes. Fractional reserve requirements for loans [was/is often 0%][1]. Also, the asset (for example) at SupplierBank is then used *as reserves to issue further debt*. Two banks in a fractional reserve system act as feedback-loop amplifiers, so if the reserve requirement was 10%, that does not mean that the money multiplier would be 10 times, but depending on flow rates and dynamic systems calculations you can end up with hundreds of thousands times the base reserve. Incidentally, this was the keystone to the financial crisis.

>  every penny received from borrowers is a pure profit. Even if Merry returned
> only 10% of her loan, the Bank still profits 10,000 because the given
> 100,000$ to Merry were not bank's money in the first place.

No. See above. The asset position is unchanged. 

> This leads me to the question: Why do banks put interest on top of
> this fraud? And what prevents banks from putting huge amount of money
> to some people's account (for example to the Bank's CEO )? This black
> magic is not quite clear to me.

Interest rate is primarily determined by interbank lending rates to cover reserve and capital requirements. 

I don't know if it applies to the rest of the world, but the main source of income for banks where I live is actually transaction fees. 

Banks could be pure digital banks that have a employees in the hundreds. Typically they are dinosaur institutions that employ in the 10s or 100s of thousands of people. The banks use fees and interest to cover the cost of employees mainly. Incidentally, this is one reason why governments chose to assist banks in 2008. Directly and indirectly they employ a big chunk of the economy, a consequence of Western deindustrialization and financialization.


  [1]: https://federalreserve.gov/monetarypolicy/reservereq.htm