Fix how Prism parses encoding from shebang args (#!)

[amomchilov]

Fix the first of the two assertion failures here:

TestRubyOptions#test_shebang [/Users/alex/src/github.com/Ruby/ruby/test/ruby/test_rubyoptions.rb:514]:
pid 49518 exit 0.

1. [1/2] Assertion for "stdout"
   | <["\"あ\""]> expected but was
   | <["\"\\u3042\""]>.

2. [2/2] Assertion for "stderr"
   | Expected /shebang line ending with \\r/ to match "".

Finished tests in 0.030572s, 32.7097 tests/s, 359.8064 assertions/s.
1 tests, 11 assertions, 1 failures, 0 errors, 0 skips

ruby -v: ruby 3.4.0dev (2024-08-22T14:01:55Z master 56a34b5af5) +PRISM [arm64-darwin23]
make: *** [yes-test-all] Error 1

(The latter was fixed in ruby/prism#3007)

Fixes ruby/prism#2996

[amomchilov]

This is the wrong place to set this, but just a proof-of-concept

[amomchilov]

okok so this is a weird one.

When not using Prism, opt->ext.enc.name is set by proc_K_option(), but not index:

ruby/ruby.c

Lines 1322 to 1324 in efa4ec0

	opt->src.enc.name = rb_str_new2(enc_name);
	if (!opt->ext.enc.name)
	opt->ext.enc.name = opt->src.enc.name;

Then, later on in load_file_internal(), the index is updated from the name, but only if no_ext_enc is true:

ruby/ruby.c

Lines 2703 to 2705 in efa4ec0

	if (no_ext_enc && opt->ext.enc.name) {
	opt->ext.enc.index = opt_enc_index(opt->ext.enc.name);
	}

I tried modifying proc_K_option() to set both of these, but that doesn't work because it can't access the no_ext_enc variable to only set it conditionally.

This split source of truth (and that they aren't always perfectly in sync) is pretty confusing!

[amomchilov]

Copied the no_src_enc/no_ext_enc/no_int_enc logic exactly as-is from load_file_internal(). Would be nice to eventually clean up this split source of truth.