Advantages of nominatim over manual geocoding ?

You seem to be under the impression that every address is present in OSM as one record, with the columns "house number", "street", "post code", "city", "country" nicely filled out. (Or to be precise, the addr:* columns.)

This is not the case.

First of all, only a fraction of existing houses/addresses actually are in OSM. Second, only roughly a third of these have a house number mapped, and even fewer have addr:street or addr:city.

Therefore your "manual geocoding" would be much more than "select coordinates from table where city='a' and street='b' and housenumber='c'". You'd have to make sure those objects that only have house numbers are still matched to the nearest street, and those that don't have a city or post code specification are matched to whatever polygon they are in (and if there's not even a city polygon but just some place=city node three kilometres away, maybe that's the city for this address). You'd also have to evaluate interpolation ranges, where OSM just has a range of house numbers and not individual numbers.

You can do all this, but you would essentially be re-inventing Nominatim.

Hence the suggestion to simply use it. A web server is not required, you can call the PHP code from the command line, or even adapt the PHP code to read input from a file.