You're looking at a device which is very far from 20mA; note SW current limit 1.5A typical (with who knows what error bars..). So yes, 100uH would be rather high for it.
But also, what's wrong with 100uH? You can get small chip inductors of much larger values, for such currents (10s mA). I'd be surprised if you need more than a 1206 (or up to 1212 size chip, inductors are oddball) for such a value, at high Q (reasonably low DCR / good efficiency).
The important things to look for is current mode operation.
Page 5 says "peak current mode". This agrees with the diagram, which shows sense current being added to oscillator ramp (slope compensation), then compared to error voltage. Control logic will be a flip-flop, and some gates to implement shutdown, soft start, blanking, etc.
Peak current mode is usable down to a modest ripple fraction, typically over 30%. This is permitted by slope compensation, which comes with the downside of the peak current varying with pulse width. So instead of a reasonably sharp current limit, as voltage ratio goes up or supply voltage goes down, the current varies somewhat proportionally. The range of current limit corresponds to the allowable ripple fraction, which is why below 30% or so is unreasonable (the spread in current limiting would vary ~3:1, being essentially useless as the low end would be too little current, while the top end would be destructive).
Without slope compensation, the minimum ripple fraction is 100%, i.e., the current rises to peak, then falls to zero, every cycle. This covers the span of DCM to BCM (discontinuous to boundary conduction mode, referring to time spent where the inductor current is above zero).
Hah, they show compensation impedance back to FB, which if they truly are doing it that way, well that proves a point I considered in some other recent thread (I already forget which); just to say that, sometimes, you might find it useful to drive the FB pin in unconventional ways, well you'll need to consider how the pin responds. In this case, the block diagram is suggesting it will have a low impedance (virtual ground) response.
Anyway, for self-contained regulators, you cannot control the value of the current sense resistor, or compensation R+C; you must use it at the recommended values, and that's that. Keep similar Thevenin equivalent values to R1 and R2 (i.e., around 13kohms, probably give or take ~2x), and you can probably use a little bigger inductor at the low voltage ratio you're looking at.
You'll probably want to experiment with exact values (resistor divider, inductor, output cap), testing transient load response (go from 50 to 100% nominal load, say by switching a resistor with a 55 timer + MOSFET).
Note they suggest mere 1uF caps, but 1uF at 1.5A for 300ns changes by 5 whole volts. You'll need larger capacitors than this. I think their interest is more to have a minimum of this much, as close to the regulator as possible; more, at up to a modest distance, will be required in total.
Tim