feat: modular Bitcoin backend with adapter pattern

[Kukks]

Summary

• Introduces a BitcoinLib interface (src/bitcoinLib.ts) that abstracts all Bitcoin operations (payments, script, crypto, PSBT, transaction parsing) behind a pluggable adapter layer
• Ships a bitcoinjs-lib adapter (src/adapters/bitcoinjs.ts) as the default backend
• DescriptorsFactory accepts either TinySecp256k1Interface (backward compatible) or a full BitcoinLib adapter (new modular path)
• All leaf modules (miniscript, tapMiniscript, ledger, signers, etc.) receive operations via parameter injection rather than importing from bitcoinjs-lib directly
• Exports PsbtLike, BitcoinLib, Network, Payment, and other adapter types from the package index

This is a non-breaking change: existing code that passes ecc to DescriptorsFactory(ecc) continues to work identically (bitcoinjs-lib adapter is loaded lazily). New code can pass a full BitcoinLib adapter to use an alternative backend.

Test plan

• All 2274 existing unit tests pass (verified locally)
• CI passes on all platforms
• Verify backward compatibility: DescriptorsFactory(ecc) works identically to before
• Verify new path: DescriptorsFactory(createBitcoinjsLib(ecc)) produces same results

[landabaso]

Hey @Kukks , thanks for the PR!!

It's quite large and I still do manual close checks for BitcoinerLab. Before I dig into it properly, let me know if you consider the PR finished on your side.

I get the idea and agree with the general direction, though I might adjust a few things or even take a slightly different approach.

Rather than dragging you through a long review cycle here, I think it's better if I take it from here once you're done. I might add more comments after I look into it further.

[Kukks]

I'm going to finish wiring up a test matrix and some accompanying fixes but yeah, feel free to take over then.

The way I've done it is to include everything in one package, but allow libs to install which peer dependency they want and treeshaking does the rest as needed

[Kukks]

Hey @Kukks , thanks for the PR!!

It's quite large and I still do manual close checks for BitcoinerLab. Before I dig into it properly, let me know if you consider the PR finished on your side.

I get the idea and agree with the general direction, though I might adjust a few things or even take a slightly different approach.

Rather than dragging you through a long review cycle here, I think it's better if I take it from here once you're done. I might add more comments after I look into it further.

It's all yours now

[landabaso]

yeah, i'm on it. btw, which repo is importing your fork so i can take a look?

[landabaso]

PR continuation notes

• I reduced differences wrt main where possible. A lot of the previous refactor pushed new code too far into the core and tests. I moved that complexity toward the adapters and kept the original descriptor / PSBT flow as intact as possible. I prefer to keep that original code as unchanged as possible.
• I fixed several typing/lint issues:
  • the any keyword should not be used in this codebase,
  • ts-ignore / linter suppressions should not be used to hide real typing problems.
• I created a small internal src/crypto.ts module instead of passing crypto functions around or depending on backend-specific wrappers.
• I restored bitcoinjs compatibility for the PSBT surface (to fix a breaking change):
  • the internal Psbt type is again an actual subset of bitcoinjs-lib Psbt,
  • native bitcoinjs Psbt instances can be passed directly to updatePsbtAsInput / updatePsbtAsOutput and the signer APIs,
  • wrapper-only methods like getInput / getTxInput were removed from the public contract and the core now uses data.inputs / txInputs again.
• I simplified the public adapter surface:
  • centralized shared network definitions/defaults,
  • removed some adapter return values that were only really there for tests,
  • restored old defaults that had changed unnecessarily (for example optional network behavior in helper paths). This fixed a breaking change.
• For Ledger, I removed TransactionOps from the public signer API. Ledger now loads bitcoinjs-lib internally for the transaction parsing it needs. Since Ledger support is optional and Ledger itself already depends on bitcoinjs-lib, this felt like the least harmful simplification. This also restored the previous signer API and fixed a breaking change introduced by the refactor.
• I internalized the bip174 types/helpers into a local src/bip174.ts, so the library no longer imports bip174.
• I fixed a number of real scure adapter bugs/parity issues:
  • txid byte-order conversion,
  • taproot PSBT metadata conversion,
  • p2tr payment parity,
  • p2sh / p2wsh input / witness parity,
  • and some subtle wrapping / interop issues.
• I modified the scure-specific PSBT interop path:
  • createScureLib(ecc).Psbt
  • createScureLib(ecc).Psbt.fromTransaction(nativeTx)
  • psbt.raw to access the native scure btc.Transaction
• I also updated the tests heavily:
  • converted the old scure test to TypeScript,
  • fixed the Jest ESM issue for noble/scure deps,
  • routed backend-agnostic tests through the backend helper,
  • and added a dedicated scure integration flow/example test.

[landabaso]

That said, I'm still not fully happy with the abstraction. The current scure flow still depends on wrapping the native scure transaction. I'd like to push that further so the scure/noble path feels more native/equivalent instead of just "bitcoinjs-shaped through wrappers".

So I do not want to merge this yet. I'd rather continue from a follow-up PR and then merge the work from both PRs together.

For now, though, this PR makes @scure/btc-signer usable and integrates it properly into the library.