XPath silently ignores erroneous evaluation where a nodeset is required

In XPath evaluation, some evaluated part is required to be a nodeset. If not, it should result in error.
https://github.com/ruby/rexml/pull/342#discussion_r3566072520

Examples:

# Function argument that requires nodeset
local-name(1)
sum("string")
count(1 < 2)

# Path after function that doesn't return a nodeset
# id("a")/path id("a")[predicate] is valid
string(a/b)[predicate]
string(a/b)/path

# Path after non-nodeset variable
$x[predicate]
$x/path

# Path after literal
"string"[predicate]
42/path

# Path after neg
-number(a/b)[predicate]
-number(a/b)/path

# Path after parenthesized expression
(1)[predicate]
(1)/path

Start by tracing REXML's XPath evaluation for the listed expressions and the discussion linked in pull request #342. Check how function arguments, path expressions, variables, literals, negation, and parenthesized expressions handle non-nodeset values. Done means each example that requires a nodeset reports an evaluation error instead of being silently ignored, while the valid id() cases remain valid.