You are of course correct that TECs do not like shock stress, so that "simmerstat" switching is indeed undesirable and PWM is much better. The frequency (several hundred Hertz) is such that thermal shock is not a concern.
So what is the problem with PWM?
Well, Peltier devices are essentially semiconductors like LEDs - they have a threshold voltage and are essentially current driven as they tend to maintain that threshold voltage. In effect, they behave as a voltage "sink" which performs their "work" in series with an internal resistance which is simply a power wastage and of course, interferes with their cooling effect (but of course, benefits somewhat, their heating effect). This means that feeding more voltage disproportionately increases this energy loss, so driving more current for less of the time, is indeed, less efficient.
What you are looking for is a constant-current driver. A simple inductor in series with your PWM driving FET, and a flyback diode across the combination, will suffice. It is not a module as such unless you obtain a complete constant-current driver at your required specification. Someone else may care to suggest appropriate parts for the inductor and diode. 
MarkT:
Peltiers are more efficient with constant voltage/current drive than PWM: Peltier Element EfficiencyThe device the OP was flailing for is a controllable DC-DC converter (which could be an LC filter
after a PWM power stage, like class-D amp)
Oh, so I needn't have bothered writing all this! 