test: type check the test tree and fix fixture drift

[robwoodgate]

Description

Nothing has ever type checked test/. Vitest strips types via esbuild without checking them, and test/tsconfig.json existed only to give eslint's typed rules a project covering the test tree, which catches rule violations rather than assignability. src/ was covered all along, indirectly: api:check runs vite-plugin-dts, which type checks it.

The result was 89 errors across 12 files, accumulated silently as the library types moved. The great majority is one field: method became required on MintQuoteBaseResponse and MeltQuoteBaseResponse, and mint quotes also gained amount_paid / amount_issued / updated_at.

Changes

• npm run check-types runs tsc -p test/tsconfig.json. The tests import the library source directly rather than the built package, so tsc follows those imports and pulls all 87 files in src/ into the same program. One command therefore covers both trees, and there is no need for a second project for src/. test/tsconfig.json gained rootDir and noEmit so a bare tsc -p behaves the same on the command line as it already did in an editor.
• test/consumer/ is excluded. Those harnesses deliberately import the package by its published name to simulate an external consumer, so they only resolve after a build. They are covered by consumers.yml, which builds first, and eslint skips them for the same reason.
• Wired into prtasks, a types.yml workflow following the existing lint.yml shape, and the pre-push hook. A full pass takes about 3.5 seconds.
• Fixtures realigned with the current types. Unpaid quotes get 0/0 accounting, which validateMintQuoteAvailableAmount short-circuits on, so those tests take the same runtime path as before. Deliberately malformed fixtures kept their hostile shape behind an explicit cast rather than being "fixed".
• Fixed the pre-push condition. It read if ! npm run check-lint && npm run check-format, where ! binds only to the first command. It aborted only when lint failed and format passed: when lint passed it short-circuited and format never ran, and when both failed it did nothing.
• AGENTS-CONTRIBUTING.md said Proof.amount is bigint and to construct with Amount.from(x).toBigInt(). That was correct when written but has been wrong since refactor: replace raw bigint with Amount VO on Proof and BlindedMessage #609 moved it to Amount, and it is where the payment request fixtures came from.

Reviewer Notes

• No files under src/ are touched and etc/cashu-ts.api.md is unchanged, so there is no public API movement here.
• The obvious alternative was shared fixture factories, so that a future field addition would need fixing in one place instead of 47. I left them out on purpose: with the gate in place, the next type change breaks the build immediately, at however many sites exist that day, rather than piling up unnoticed until someone opens the file. Fixing one or two sites as they break is cheaper than maintaining factories, and the factories would not have prevented today's cleanup anyway.
• The gate was checked for teeth rather than assumed: planting a Proof with a raw bigint amount makes check-types exit 2 and the hook exit 1, and both return to 0 once removed.
• The sweep also surfaced several src/ defects where a declared type asserts more than the code delivers. Those are deliberately not in this PR; they follow in a separate one stacked on this branch.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.76%. Comparing base (20a29e2) to head (ced752f).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #954   +/-   ##
=======================================
  Coverage   95.76%   95.76%           
=======================================
  Files          55       55           
  Lines        5781     5781           
  Branches     1465     1465           
=======================================
  Hits         5536     5536           
  Misses        104      104           
  Partials      141      141           

Flag	Coverage Δ
integration	37.75% <ø> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.