Fix Node process crash in compressFileWithGZIP on unreadable PUT source

[kartikgupta2607]

Description

FileUtil.compressFileWithGZIP (lib/file_util.js) pipes the source file through gzip into a temp file and awaits a promise that only listens for the writable's finish event — none of the read/gzip/write streams has an error handler. If the source file is missing or unreadable when a PUT ... AUTO_COMPRESS=TRUE runs, the read stream emits an error with no listener, which Node escalates to an uncaughtException that crashes the whole process; the wrapping promise also never settles.

This is reachable whenever a file disappears between when the PUT file list is resolved and when a given file is compressed (an external cleanup process, a race, or a path that no longer exists).

Fix: use stream.pipeline from node:stream/promises. It propagates an error from any stream in the chain into a single rejection and destroys every stream, so a missing/unreadable source now rejects compressFileWithGZIP — which propagates through the existing try/catch in uploadOneFile and surfaces as a failed PUT — instead of terminating the process. No behavior change on the success path.

Minimal repro (crashes on current master):

const os = require('os');
const { FileUtil } = require('snowflake-sdk/lib/file_util');
await new FileUtil().compressFileWithGZIP('/path/does-not-exist.csv', os.tmpdir());
// -> uncaughtException: ENOENT (process crash); the promise never settles

Checklist

• Create tests which fail without the change (added FileUtil.compressFileWithGZIP() tests: gzip round-trip + rejection when the source file is missing; the latter crashes the runner without this fix)
• Unit tests pass (npm run test:unit); prettier, oxlint and check-ts pass. Integration tests not run locally (require a live account)
• Extend the types in index.d.ts file (not necessary — no public API change)
• Updated CHANGELOG.md
• Provide JIRA/GitHub issue id (external contribution; no JIRA id — happy to link a GitHub issue if preferred)

Draft: opening for early feedback.