a) From reading all those topics it seems that the answer depends mostly on the design of the particular LDO, specifically whether the regulating element is FET or BJT. Not all datasheets have internal schematics available.
b) Most of the modules on the market have 5 to 3.3V LDOs nowadays. Not all of them are adequately designed. Case in point: in our product we are using Adafruit OLED display, which has on board 150 mA MIC5225 LDO for a panel that consumes up to 200 mA with all pixels white.
c) Pure 3.3V systems are now ubiquitous and they often have power management (e.g. via EN input on LDO or DC-DC) implemented. Adding separate 5V switch for the modules that use 3.3V internally anyway seems to be bad idea.
So, here are my questions:
Is it safe to permanently apply 3.3V to the output of MIC5225, which has IN and EN pins connected together but floating otherwise? The datasheet has examples for IN floating/EN grounded and for both grounded. In either case the "reverse leakage protection" seems to be limiting current to about 5 µA.
Would grounding both IN and EN (as in datasheet) be better than leaving them unconnected? Just as in previous question, when I say "safe" or "better" I mean no magic smoke, additional battery drain or stuff like that.
Is there universally suitable way to bypass unnecessary LDO on the pre-made modules (assuming there is an access to 3.3V output but no easy way to modify schematics)? This is obviously quite common problem (see the links in the beginning), so I think it deserves a definitive answer.
For the latter the opinions are rather conflicting. Here are some of them:
Let it be. The worst that can happen is capacitor on the input will be charged to output voltage;
Add bypass diode / ideal diode / MOSFET if LDO has BJT output, otherwise see (1);
Short-circuit input to output. I must admit, this one seems most compelling to me, with a benefit of re-using input capacitance for decoupling. However (at least in my case) this could be bad, since EN pin becomes powered too.