See discussion in #432 (comment):
@eggrobin:

[…] the question is whether this should be inverted.

        propSource = getProperties(version);  // ToolUnicodePropertySource.
        property = propSource.getProperty(xPropertyName); 
        if (property == null) { 
            propSource = getIndexedProperties(version);   // IndexUnicodeProperties.
            property = propSource.getProperty(xPropertyName); 

@macchiati:

I agree, it should be inverted. There are a few constructed properties
(non-UCD) that are only available via Tool... That is, the extra
normalization properties. So those should fallback to the Tool

This uses IndexUnicodeProperties except for the following, where ToolUnicodePropertySource is used:

1. the toNF[ ] pseudoproperties;
2. properties that did not exist in that version of Unicode, if the version is requested with an R (for Retroactive) rather than a U.

This is used for the invariant about NFKC_SCF, which does not exist until 15.1, but whose derivation would be stable very far back.

This also fixes defaults (incorrectly null instead of <code point>) in the NFKC_CF and NFKC_SCF returned by ToolUnicodePropertySource.