A real continuous slider can't really exist in the digital world (vs analogical), as it's impossible to consider all the real numbers lying in an arbitrary range nor to trigger events for all possible transitions. In the end it always relies on a range of discrete values.
Still, you could improve the precision of the slider by incrementing the number of steps in the given range, for example if the precision has to be in percent, one obviously needs 100 values between 0 and 1 :
nsteps = 100
slider = { 'steps': [{ 'value': step/nsteps } for step in range(nsteps + 1)] }
Example with values in the range [ 0.1, 10 ] and a higher precision, using numpy.linspace for convenience :
nsteps = 10**5
slider = { 'steps': [{ 'value': step } for step in np.linspace(0.1, 10, nsteps)] }
Not sure how the front-end will behave with very high precision though, it must depend on how changes are detected, how the corresponding events are emitted by Plotly.js (at which rate), and above all what happens in the handler.