fix(cruby): SAX and Push parser error handling in the presence of foreign handlers

[flavorjones]

What problem is this PR intended to solve?

When libxml-ruby is loaded before Nokogiri, it sets its own global libxml2 error handlers. There are a few places where Nokogiri does not defensively remove that handler or use its own handlers. Commit 771164d (shipped in v1.11.0) removed one instance of a defensive removal, and that combined with the presence of libxml-ruby caused some tests to fail in ActiveSupport's test suite.

• [bug] nokogiri-1.11.0-x86_64-linux causes Active Support failure #2168
• NokogiriSAXEngineTest#test_exception_thrown_on_expansion_attack fails with nokogiri v1.11.0 native gems rails/rails#41015

This changeset introduces a systemic approach to testing handlers, setting a foreign handlers before every test in the suite and asserting that it's never called. It then introduces defensive calls to xmlSetStructuredErrorFunc where necessary.

Notably, it also introduces a new pattern of save-and-restore the previous handler in one place where we're recursively calling HTML parsers (see EncodingReader for details). This is a pattern we should extend to all other places in the code, but in the interest of time I'm not attempting that in this changeset (I'll open a new issue to track that work and backlink to this).

Have you included adequate test coverage?

Yes, a significant part of this changeset is introducing additional test coverage.

Does this change affect the behavior of either the C or the Java implementations?

This changes the behavior of the C extension only, as it relates to libxml2 behavior.

[codeclimate]

Code Climate has analyzed commit f9a2c4e and detected 0 issues on this pull request.

The test coverage on the diff in this pull request is 100.0% (80% is the threshold).

This pull request will bring the total coverage in the repository to 94.3%.

View more on Code Climate.