Merge upstream v2 (2.19.6 → 2.20.5)

[grantcox]

Brings the fork up to date with upstream v2 (2.19.6 → 2.20.5, 65 commits). Here is the upstream changelog: https://github.com/recurly/recurly-client-ruby/blob/v2/CHANGELOG.md

Conflicts are resolved in upstream's favour wherever the two sides are equivalent, so that diffing this fork against upstream surfaces only the differences that actually matter.

Commits

1. Merge upstream/v2 — 76 files, +2770/−112. Conflicts resolved:
   • lib/recurly_v2/xml/rexml.rb — took upstream's &Proc.new { |e| yield e } over our equivalent &block form. Verified on Ruby 4.0.5 that the two produce identical traversal order; they differ only in the error raised when called with no block, and XML.filter (the sole caller) always passes one. This file now matches upstream exactly apart from the module name.
   • spec/recurly_v2/api/errors_spec.rb — the conflicting line was only Recurly::API vs RecurlyV2::API, resolved to the renamed form. Upstream's MiniTest::Mock → Minitest::Mock fix was identical to the one in Fix CI, drop the V2 label requirement, and test through Ruby 4.0.5 #5 and auto-merged.
   • spec/recurly_v2/item_spec.rb — took upstream's new GL-account assertions.
   • lib/recurly_v2.rb — took upstream's three new requires.
   • Seven files upstream added under lib/recurly/ and spec/recurly/ were relocated into the renamed _v2 directories.
2. Apply the RecurlyV2 rename to the code upstream added, following 24843a2: Recurly → RecurlyV2, recurly/ require paths → recurly_v2/.
3. Stop expecting nil via must_equal in two specs upstream added — minitest 6, which Ruby 4.0 bundles, rejects it.

Remaining intentional divergence from upstream

lib/recurly_v2/xml/nokogiri.rb keeps our &block fix. Upstream still has the bare &Proc.new, which Ruby 3.0 removed, so that backend raises ArgumentError on every modern Ruby upstream supports. This one is a real fix, not a stylistic difference.

What upstream adds

Almost entirely additive API surface: external payment phases, General Ledger Accounts, performance obligations, rev rec, business entities, taxable address sources, card_network_preference, net_terms_type, proflex options, invoice and line-item refunds, external subscription lookup by uuid/external_id, vertex_transaction_type, gift card tax_service_opt_out, action_result, and ramp interval starting_on/ending_on.

No behavioural change to anything the fork already used.

Verification

All 13 matrix jobs green (2.3 → 4.0.5, latest 4.x, jruby-9.3).
Suite grew 546 → 604 runs, 1178 assertions, 0 failures.
XML.filter exercised directly on Ruby 4.0.5 against both backends — card numbers and verification values mask correctly.

[grantcox]

@becky-ynab I'm on the fence about whether we do this update. On one hand, I'm not aware of any security related patches in the upstream versions, so there's no real need to update. And why touch a production dependency, that we want to remove in the near/medium future anyway, if not necessary?

But on the other hand, it does feel like a good practice to not be 4 years out-of-date on our dependencies, especially with our payment processor.

The corresponding Evergreen PR (using this bumped / merged version) is https://github.com/ynab/evergreen/pull/30124 , and everything is green there. But how much our automated tests are actually relying on this gem, vs just mocking it out, I can't say.

[becky-ynab]

👍 thank you!