In response to comments
No, when inputting the priority fees you want to use into the API request on Jupiter, that will be the amount you spend for the priority fees in total. When adding the priority fees "manually" as an instruction you should be aware of difference between the compute unit price and compute unit limit. Lets first talk about what a "compute unit" is. A compute unit is kind of a measurement of how much computation you need in order to process your instructions. A single Solana transaction can have multiple instructions (luckily Jupiter transactions are normally only a single big instruction).
By default, each instruction within a transaction has a compute unit limit of 200,000. This can be changed (to a max of 1.4M) using the set_compute_unit_limit(). But beware that a higher compute unit limit can decrease the chance that your transaction will be included on chain, since it takes more computation for a validator to process it.
In the Jupiter API request in your question the param "dynamicComputeUnitLimit" is set to True. This parameter will basically simulate the transaction and then use set_compute_unit_limit() to set the compute unit of your transaction to an appropriate amount, whereas if that was false it would be set to 1,400,000, this a default value set by Jupiter. (If you are creating a transaction and its instructions from scratch it could be a good idea to use the same method as Jupiter does where you simulate the transaction and then use that as a ballpark as to how much your compute unit limit should be, but in the scope of using Jupiter just leave that param to True).
The compute unit price is the price (in lamports) you are paying to the validator per unit of computation.
For example, if you send a transaction with a single instruction, which has a compute unit limit of 200,000 and you set your compute unit price to 2.8 lamports per unit. Then you will end up paying a priority fee of 560000 lamports which is (560000 * 10^(-9) =) 0.00056 SOL.
No, it only sets the price you are going to pay per compute unit. The function is defined in the solders python library.
Remember that if you're using the Jupiter API and you are using the "prioritizationFeeLamports" param in your API request then there is no reason to change the compute unit price, since this is already done by Jupiter. Jupiter will automatically calculate how much you need to pay per compute unit in order for the priority fee to be the amount you specified. Thus the transaction already has the compute unit price instruction present and there is no reason to add it again.
If you want to add it to a transaction which doesnt have it, follow the steps I mentioned in my initial answer, where you need to add the instruction manually.
Also Solana is having a lot of congestion recently and some transactions do fail more often than others. Maybe try setting a higher priority fee.