Fix grammar for arbitrary equality comparisons in dependency specifiers

[notatallshaw]

DPO thread: https://discuss.python.org/t/mistakes-in-dependency-specifiers-grammar/105190

If you use the example script with the inserted grammar and use the test name===1 you get the following exception:

ometa.runtime.ParseError: 
name===1
       ^
Parse error at line 1, column 7: expected EOF. trail: [letterOrDigit version version_one version_many versionspec]

This is because if == is listed before === then the first two characters of === are matched against ==. This PR changes === to be listed before ==, after which you get:

name===1 -> ('name', [], [('===', '1')], None)

---

📚 Documentation preview 📚: https://python-packaging-user-guide--1969.org.readthedocs.build/en/1969/

[notatallshaw]

Oh also this fixes the ordering of <= and < in the version specifier grammar which is incorrect and inconsistent with the ordering listed in the full grammar file.

I copy and pasted the correct ordering from the full grammar file, so I didn't notice this fix until rereading the PR.

[ncoghlan]

Merging as a bug fix (since any parser that actually worked couldn't have been using the grammar as written)