`redundant-object` incorrectly counts references within recursive formations, causing performance issues

[yegor256]

misc/redundant-object.xsl:16 reports an object as redundant when
count(key('referenced-by-name', @name)) <= 1, that is, when the whole
document mentions its name once. The count is purely syntactic, so it treats a
lone reference sitting inside a recursive formation the same as a lone
reference in straight-line code.

Those two are not the same. Take eo-runtime/src/main/eo/number/arc-sine.eo:45
in objectionary/eo, where point.times point > squared is declared beside the
recursive helper poly and referenced exactly once, from inside poly. The
lint says it "may be inlined". Following that advice is a mistake: poly
recurses thirty levels, and while the named attribute gives all thirty levels
one shared series.squared node, the inlined form gives each level its own copy
of the point.times point subgraph nested inside the previous level. Running
the eo-runtime suite with the object inlined takes EOarc_sineTest from about
9 seconds to 189, and under the parallel suite the JVM exhausts its heap.

The smallest fix is to skip the defect when the single reference lies inside a
formation that is an ancestor-or-self of a recursive one — that is, when the
referring node has an ancestor o whose own name appears in the @base of one
of its descendants. This is not the same gap as #639 and #640, which are about
references the lint fails to see at all; here the reference is counted
correctly, but one syntactic mention is many dynamic ones.

[yegor256]

@volodya-lombrozo could you take a look when you have a moment? Happy to clarify anything in the report that is unclear.

[0crat]

@yegor256 Thanks for closing this issue! You've earned +12 points for this: +8 as a basis; +4 for resolving it in less than 12 hours. Your running score is +2757; don't forget to check your Zerocracy account too).