The GeoTIFF "value-added" products have both the ProbabilityLiquidPhase and a split-out of the precipitation into liquid and solid (note—both are liquid equivalent). The documentation is at https://pps.gsfc.nasa.gov/Documents/README.GIS.pdf and quoting from p. 14-15:

If one is interested in the phase of the precipitation, one should download the compressed file with the *.zip extension. The .zip file contains GeoTIFF and WorldFiles (.tif and *.tfw files) for total precipitation (TP), liquid/mixed precipitation (LP), liquid-equivalent ice-phase-only precipitation (IP), and the percent of liquid/mixed-phase vs. total precipitation (PERCENT). 15 In the following discussion, TP, LP, and IP should be taken to have units of millimeters if one is discussing real-time or research values for precipitation accumulation during the period of an IMERG GIS file. In contrast, if one is working with the research values for average precipitation rate during the period of the IMERG GIS file, then the values if TP, LP, and IP should be taken to be millimeters per hour. To make the four IMERG GIS variables consistent, the following relationships always hold true. The total-precipitation file is always equal to the sum of the liquid/mixed-phase precipitation file and the liquid-equivalent ice-phase precipitation file. The liquid-precipitation percent field is always 100 times the ratio of the liquid/mixed-precipitation field and the total precipitation field. These two relationships are expressed in Equation 1.

The rules for how to create the time-average probabilities are:

1. At the native half-hourly time interval, the probability is based on a specification relationship for rain vs. snow. Mid-range probabilities are unlikely to indicate "mixed" solid and liquid, which is a rare event, except perhaps over the Southern Ocean. It just means you have to pick a threshold and accept a fair amount of uncertainty. For averages shorter than a day, the same concept holds. For averages of a day or more, we think that there are enough samples that the probability should represent the (precipitation amount-weighted) fraction of the total that fell as liquid, with the rest falling as solid.
2. The default suggested threshold is 50%, but for a specific region you could choose something else if you have a basis for the choice. This is why we give the probability, not just a categorical liquid/sold statement.
   Snowfall amounts are deficient in Version 07, and precipitation in general is less certain in mountainous terrain.
3. So, IMERG is likely not a good source for estimating snowpack, but this discrimination could say for any given storm whether or not the precipitation is falling as liquid, and therefore immediately available to drive flooding and landslides.