feat: complete listen key renewal support

[AIVensk]

Closes #24.

Changes

• Add ListenKeyService.extendListenKey(listenKey, account) using the existing PUT endpoint.
• Export BingxExtendListenKeyEndpoint through the public package entry point. Use direct imports inside the endpoint to avoid a circular import when loading that entry point.
• Add nine regression tests for the public export, method/path/parameters, API key and signature, service completion/error propagation, and existing key generation.
• Mark listen-key renewal as implemented in the README.

Verification

Passed locally:

npm test -- --runInBand --runTestsByPath src/bingx/endpoints/bingx-extend-listen-key-endpoint.spec.ts src/bingx-client/services/listen-key.service.spec.ts src/bingx-client/bingx-api.client.spec.ts src/nest-bingx/nest-bingx.module.spec.ts
npx --no-install tsc --noEmit --incremental false
npm run build
npx --no-install eslint src/bingx-client/services/listen-key.service.ts src/bingx-client/services/listen-key.service.spec.ts src/bingx/endpoints/index.ts src/bingx/endpoints/bingx-extend-listen-key-endpoint.ts src/bingx/endpoints/bingx-extend-listen-key-endpoint.spec.ts
git diff --check

The focused run passes 18 tests across four suites, including all nine new tests. The tests use synthetic credentials and mocked executors.

Existing baseline failures

I also compared the broader local WebSocket/unit suite against clean upstream e10420ed2294be869424088f5eb699ecff6e4a82, excluding the credential-dependent exchange playground:

timeout 35s npm test -- --runInBand --testPathIgnorePatterns=/playground/

• This branch: 31 passed, two failed.
• Unchanged upstream: 22 passed, the same two failed.
• Both runs encounter the account-WebSocket reconnect timeout and order-push undefined-socket failures, then lingering handles and the 35-second timeout.
• Whole-project ESLint reports the same existing unused OrderTypeEnum import in src/bingx/interfaces/websocket-event.ts:1 on both revisions. All changed TypeScript files pass lint.

Implementation and review used AI assistance; the checks above were actually executed.