i skipped the detail for simplicity,
Well, there is your problem. DO NOT SKIP details. I don't know you and i don't know your proficiency level in all this stuff. So i assume the worst case scenario : this guy i susing CFB opamps and shows schematic for a VFB opamp ... ouch ...
You also mentioned TI's parametric search so i pointed you to Digikey's parametric as it is manufacturer independent.... since i judged your profiency level low ( becasue you did not feed me all info / false info ) i unrolled the digikey walkthrough. .. So don't get pissed at me if i'm trying to help . OK ?
As for the impedance matching between the inputs : Opamps generate a biascurrent on their input pins. this current can flow in or out of the opamp depending on the internal technology used ( bimos , cmos , bipolar , jfet and where the current pump is for the longtail. ( if the pump sits to the ground side the current will flow in the opamp , if the pump sits against the rail the current will come out of the inputs ).
This current will create a voltage drop across the impedance of your source. Since and opamp is a DIFFERENCE amplifier : you want to create the same drop on BOTH inputs , so it cancels itself out. You are creating a common mode voltage made by the bias current of the inputs . if you the inputs see a different impedance you are no longer creating a common mode but a differential error. this will get amplified by whatever gain you have in your system already... 1 microampere may not be much , but into 1 megaohm it is 1 volt. if your opamp has a gain of 20 ... the output will be stuck against positive rail no matter what you try ...
So , yes , impedance matching is important. especially when dealing with large ohmic values.
Now, back to the problem at hand.
THe THS3092 is not suitable for what you want to do due to its large input bias current that thing is typcially 20uA and it drifts all over the place.
So here is your dilemma.
Let's say you measure 10 volts.. 10 volts into 1 megaohm is 10 uA... but, the opamp also gives 20ua ... that is 30 uA flowing through the 1K resistor. you just injected a 200% error into your measurement.. Send that through whatever gain you have and it all goes to snot. Let's say you can calibrate that away .... Heat it up or cool it down and it still drifts all over the place.
in other words you need something that is CMOS input based , or has an input bias current in the order of picoampere. A quick look found nothing beneath 75nA and those are all 5 volt max..
the only one i can find that fits the 30 volt rail criteria is the LM6172... and that has already a 1uA input bias...
so if you measure 1 volt at the input you already have a 100% error ...
You may have to drop that 1 megaohm down .... and live with a lower input impedance.
besides that 1 megaohm is going to be annoying when you start going above a few MHz.... you will have to put a small ( few pf ) cap in parallel.. that resistor is going to become inductive... unless you use thin film types. and even then.... the stray capacitance on your board is going to kill you..
1pF at 10 Mhz is roughly 15K... in parallel with your 1k ... thats a 7% error. and that isassuming your stray capacitance is 1pf ...
it'll be more in the order of 5 pf... and then you have rouglhy 3K ... in parallel with 1k... that's a big error ....
so you will need to compensate that with a capacitor across the 1 Mohm to balance it out.