T-16672 Throw on ingesting errors so AWS retries the invocation

[PetrHeinz]

During a Better Stack ingest outage, logs forwarded by this Lambda were silently lost: @logtail/node retries a failed sync 3 times, then drops the batch, console.errors, and resolves log() as success — so the invocation succeeds and AWS never retries.

This sets throwExceptions: true on the Logtail client, so a delivery failure (after the client's own retries) rejects the handler's Promise.all and fails the invocation. CloudWatch invokes the Lambda asynchronously, so AWS redelivers the batch with its default async-invoke retry policy — no setup needed from customers.

• Delivery becomes at-least-once: if part of a batch was delivered before the failure, the AWS retry re-sends the whole batch, so duplicates are possible during flaky periods.
• Added a test asserting the handler rejects when the ingest endpoint returns 500 (uses a cache-busting import, since the logger is built from env at module load).