test: expand offline descriptor test coverage to include all BIP descriptor constructors

从 tests/test_offline_descriptor.py 和现有的 test_descriptor_bip86 测试开始，遵循其 fixed-mnemonic 和 full-descriptor-string 模式。为列出的 BIP44、BIP49、BIP84、BIP86、new_wsh_sortedmulti 和 new_sh_sortedmulti 构造函数添加覆盖，然后运行离线 descriptor 测试；在无需网络访问的情况下，所有预期的 descriptor 字符串和校验和都通过后即完成。

What is missing

tests/test_offline_descriptor.py currently contains a single test:

test_descriptor_bip86 — tests Descriptor.new_bip86() for the external keychain only

bdkpython exposes 8 BIP descriptor template constructors (from bdk_wallet::template)
across 4 standards, each with a private key variant (DescriptorSecretKey) and a
watch-only public key variant (DescriptorPublicKey), plus 3 miniscript multisig
constructors. None of the following have test coverage today:

What I propose to add

13 new tests in tests/test_offline_descriptor.py covering:

• BIP44 — external keychain, internal keychain, and watch-only (new_bip44_public) variant
• BIP49 — external keychain, internal keychain, and watch-only (new_bip49_public) variant
• BIP84 — external keychain, internal keychain, and watch-only (new_bip84_public) variant
• BIP86 — internal keychain and watch-only (new_bip86_public) variant (external already exists)
• Multisig — new_wsh_sortedmulti and new_sh_sortedmulti 2-of-3

I am intentionally leaving new_sh_wsh_sortedmulti out of this PR to keep the scope
focused, and can follow up with a separate issue if the maintainers consider it worthwhile.

All tests will be:

• Fully offline — no network access required
• Deterministic — fixed test mnemonic so expected outputs are independently verifiable
  with any compatible BIP32/BIP44/BIP49/BIP84/BIP86 implementation
• Following the same pattern as the existing test: fixed mnemonic → derive key → assert full descriptor string including checksum

Why this matters

Descriptor construction is the first thing every bdkpython user does — it is required
to create a Wallet. Without test coverage for these constructors, a regression in the
FFI layer (e.g. wrong derivation path, wrong script type, wrong network prefix, wrong
checksum) would go undetected until a user reports it in production.

Each test pins the full descriptor output string to a known-correct value, which means
any future change to the FFI type conversion or the bdk_wallet::template behavior
that silently produces wrong output will immediately fail CI.

This is a pure Python addition — no changes to the Rust layer.

Checklist before I open a PR

• Maintainer confirms this contribution is welcome
• I will follow the existing file structure in tests/test_offline_descriptor.py
• I will verify all expected values against the fixed mnemonic before submitting