fix: return transaction objects on repeated bridge swaps

[newmattock]

Summary

• Fix Transaction.makeTxObject() so it returns a fresh Bitgesell transaction object on every call, not only during the first SDK initialization.
• Add a focused regression test proving repeated factory calls no longer return undefined.

Verification

• pnpm exec jest src/__tests__/Transaction.test.ts --runInBand
• pnpm exec tsc --noEmit --skipLibCheck --module commonjs --target es2019 --esModuleInterop --types jest src/bridge/Transaction.ts src/__tests__/Transaction.test.ts
• pnpm exec prettier --check src/bridge/Transaction.ts src/__tests__/Transaction.test.ts
• git diff --check

Notes

• pnpm exec tsc -p tsconfig.json --skipLibCheck currently fails on upstream because the repo's tsconfig.json references the node type package but @types/node is not declared as a visible dependency in this install. The focused compile command above covers the changed source and regression test.
• pnpm exec eslint src/bridge/Transaction.ts src/__tests__/Transaction.test.ts --ext .ts did not complete locally after roughly 45 seconds and was stopped.

Bounty context

Submitted for Bitgesell improvement bounty consideration: BitgesellOfficial/bitgesell#39.

[MyTH-zyxeon]

Review-assist note for the Bitgesell #81 improvement lane.

I checked the public diff and repo test config for this PR:

• Scope is narrow: Transaction.makeTxObject() now initializes jsbgl once but returns a fresh globalThis.Transaction on every call, which covers the repeated-call case instead of returning undefined after the first initialization path.
• Regression coverage is present in src/__tests__/Transaction.test.ts; the repo's Jest config uses roots: ['src/__tests__/'], so the new test should be discoverable by the normal Jest path.
• No bridge secrets, RPC endpoint changes, signing-path changes, or live transaction execution are introduced; the patch stays in the factory/test layer.

One maintainer-side check I would suggest before merge: run the repo's normal Jest suite, or at least the new transaction factory test, to confirm the mocked @naftalimurgor/jsbgl import behaves the same in CI. From the public patch shape, this looks like a small, reviewable fix.