The opamp has internal input capacitance (a few pF) which is being discharged by I1. I think TL071 datasheet fails to specify it
but it's something you expect to exist in every chip.
The reason the opamp fails to prevent it is because it has limited bandwidth and limited gain at high speeds. It takes a significant input error (a voltage spike at Vs) to cause the opamp to start moving the output, which increases current flow through R1||C1, compensates the increased I1 current and slowly discharges Vs. You can see that the output isn't exactly equal to the input current, it's delayed and distorted.
A faster opamp will do it faster. If you are using LTspice, try the built-in LT1022 model. Or heck, even LT1468 which is a significantly faster bipolar opamp. Bipolar might still be good enough with only 50kΩ feedback resistance. Maybe.
You will never eliminate it completely, unless you replace the opamp with something much more ideal, that has zero phase delay and constant open loop gain at all frequencies. Try the 'e' component (voltage-controlled voltage source). Connect its inputs as an opamp, connect the negative output to ground, set its E value to 1Meg (open loop gain).
edit
T3sl4co1l is right that some of the output distortion is due to C1. However, the distortion of combined C1||R1 current is due to limited opamp speed, otherwise this current would exactly follow I1 current. And the fact that there are spikes on Vs is a clear proof that the old '071 needs a serious kick on its butt to get it going