fix: clear errors for connection, auth, and API failures

[jpetey75]

Closes #1

Problem

A misconfigured client surfaced a raw, Lightdash-agnostic error that gave no hint at the cause:

client = Client(instance_url="https://app.lightdash.com", access_token="xxxx", project_uuid="xxxx")
client.list_models()
# httpx.ConnectError: [Errno 8] nodename nor servname provided, or not known

Change (intentionally minimal)

A nested try around just the request translates the two reported failure modes into clear, catchable exceptions. The existing raise_for_status + structured-error handling is left completely untouched (unchanged context in the diff):

Failure	Exception	Message
DNS / refused connection / timeout	LightdashConnectionError	names instance_url + likely cause
HTTP 401 / 403	LightdashAuthError	points at the access_token

Both subclass LightdashError, so existing except LightdashError handlers keep working, and the original httpx error is chained as __cause__.

Production diff is +61/−9 across 3 files; client.py is +37/−9, all additive around the request — no change to the success path or the existing API-error path.

Scope note: an earlier revision also reworked the 4xx/5xx path to surface the API's message on HTTP errors. That changed existing behaviour, so it was cut from this PR to keep it small; it can be a separate change later if wanted.

Verified live (the issue's two exact cases)

CONNECT -> LightdashConnectionError (0): Could not connect to Lightdash at
           https://this-host-does-not-exist.invalid. Check that instance_url is correct...
AUTH    -> LightdashAuthError (401): Authentication failed (HTTP 401) for
           https://analytics.lightdash.cloud. Check that your access_token is valid...

Tests

tests/test_client_errors.py (8): connect-error wrapping, timeout wrapping, LightdashError subclassing, cause-chaining, 401/403 → auth error, plus two guard tests that the existing structured-error and success paths are unchanged. Full unit suite green (131 passed).