It looks like both filter sections are designed with the -3dB point at the same frequency, or very close together, so this filter is doing what it should.
In the crossover region, both sections contribute to the output, so it is higher than either alone. The slight peax at the crossover frequency would be 3dB if both signals were in phase(so they added coherently), so presumably they aren't. EDIT : apparently the small separation between -3dB points, rather than phase, accounts for this peak being less than 3dB.
For a classic design without that bulge, read up on the Linkwitz-Riley crossover, commonly used in loudspeakers where you want HPF and LPF outputs to sum to unity.
I don't know what you were expecting but if you wanted a notch you'd have to separate the -3dB frequencies, then the depth of notch will depend how far apart they are, and it won't be a deep notch.
If you wanted a deep notch, one approach is the Twin-T filter which can be made as narrow as you want.
Or start by specifying the frequency, notch width (at -3dB), and notch depth you want, and research filter design techniques to meet that specification.
Your understanding of op-amps is fundamentally correct, and as Olin notes, an output with no load may very well drive close to the rails, but many parts will struggle even at no load.
What you may not understand is the models used for simulation, and these vary considerably in detail and accuracy.
This application note explains why most op-amp models are continuous time and why earlier models may not accurately show the limitations of the output. It also goes into some detail on how these models have evolved to bring greater accuracy to simulations.
Most interestingly, the model itself has no real relationship to the part in terms of the actual circuitry used, as the model is only representative of the behaviour of the part; these models rarely model start-up response (if ever) which can catch the unwary. (Chopper stabilised device outputs can be interesting for the first few milliseconds).
Understanding the limitations of simulation tools is critical in engineering, and only a thorough understanding of the parts being simulated (op-amps in this case) will save you from serious circuit mistakes.
The simulation is to help you understand the typical performance of a given part in a given configuration to help avoid many problems; you still need to understand the device fundamentals.
I have seen models (TVS devices in that case) that did not reflect reality and caused quite a lot of embarrassment when the box was subjected to lightning tests, because the designer had blindly believed the simulation.
Best Answer
Here is a basic circuit that will come close to your requirements.
I showed using a nice opamp from Linear Technology that can work at 5V and still allow the signal to go down near to GND. To get this to work with older, not so nice opamps like a 741, you would need to use supplies of +/-5V or +/-6V.
This shows the response of the above circuit. The output range does not go quite to the limits that you specified for your A/D converter input. It is a good idea to design with some margin to deal with variation in components and the input signal range.
Note that the above circuit inverts the range of the sensor. The inverted opamp configuration was convenient in this case but the circuit could just as well have been designed in a non-inverting manner. On the other hand it takes a trivia amount of MCU software to offset the A/D readings to remove the inversion if that proves necessary.