What happens to missing data during training when use_missing=False?

[hsorsky]

Similar to #2921, but regarding what happens if we have use_missing=False.

Apologies for opening a new issue, but that issue is "locked and limited to collaborators.". If I understood what disabling use_missing=False actually did, I'd be happy to try to improve the docs around it, but at the moment I do not understand it.

I am not well versed with C++, but from

LightGBM/src/io/bin.cpp

Lines 322 to 323 in 2db0b25

	if (!use_missing) {
	missing_type_ = MissingType::None;

and

LightGBM/include/LightGBM/tree.h

Lines 260 to 262 in cc6a2f5

	if (std::isnan(fval) && missing_type != MissingType::NaN) {
	fval = 0.0f;
	}

it looks like, even if we set zero_as_missing=False we'd still treat missing values like zeros?