Make string_value raise what StringValue raises

[youdie006]

string_value is documented as "Like StringValue in C" (lib/cgi/escape.rb:235), but #{input.class} differs from it two ways:

CGI.escapeHTML(nil)              # C:  no implicit conversion of nil into String
                                 # RB: no implicit conversion of NilClass into String
CGI.escapeHTML(BasicObject.new)  # C:  TypeError
                                 # RB: NoMethodError

The second one matters more than the message: a rescue TypeError that works with the extension does not catch without it.

lib/cgi/escape.rb:167-168 skips the C extension on TruffleRuby, so that engine always takes this path. Confirmed on truffleruby 3.3.7 — CGI.escapeHTML resolves to CGI::Escape, and both rows above reproduce.

Letting String#+ do the conversion makes core raise exactly what the extension raises. String === input returns the argument untouched for strings and subclasses, so there is no extra allocation on the common path, and it never calls a method on the object — which is what keeps BasicObject a TypeError.

The test goes in CGIEscapePureRubyTest, which removes the C methods in setup and therefore covers the fallback on every engine. #131's own nil assertions are in CGIEscapeTest, which on CRuby exercises the C path — that is why this shipped untested.

Verification

Run on truffleruby 3.3.7, the engine that always uses this path:

row	result
new test on master	1 failure
with the fix	5 assertions, 0 failures
revert	1 failure
name nil but keep input.class for the rest	fails at the true row
String.try_convert(input) || ("" + input)	4 of 5 pass, fails at BasicObject

That last row is worth a note: String.try_convert itself calls respond_to? on TruffleRuby, so it raises NoMethodError before any fallback can run. Keeping try_convert in the expression looks equivalent and is not, on the one engine this fix is for.

CRuby 3.2.11: same red/green, and the full file goes 442 → 443 tests with 0 failures. (3 errors on both pristine and fixed — RbConfig::LIMITS and assert_separately, because I ran the file directly rather than through test/lib/helper.rb.)

Behaviour table against the extension, both engines: String and String subclasses come back as the same object, to_str objects convert, and nil / true / false / Integer / BasicObject all raise the extension's exact TypeError message.

I did not run the jruby row; its escape backend is the Java extension, and CGI::Escape is only reached there when that is absent.

Unrelated to this, I have #136 open on the same file — different method, no textual overlap, and either order works.

---

Disclosure: I used Claude (an AI assistant) while preparing this change.