fix: attach callId and map fetch failures

[Arshgill01]

What

1. Wrap the Calls transport fetch so a rejected fetch becomes CalleConnectionError instead of a raw TypeError.
2. After createAndWait successfully POSTs, attach callId on CalleAPIError, CalleConnectionError, and CalleTimeoutError thrown from the wait-phase GET.
3. waitForResult timeout already named the id in the message; it now also sets the callId property.

Why

createAndWait kept call.id in a local variable. If the following GET failed (network, 5xx, deadline), the thrown error had no callId field, so the caller could not recover a call that had already been created. A rejected fetch was not mapped to CalleConnectionError, so catch (e instanceof CalleError) missed it.

Related but not fixed: #17 (wait returns on terminal status before structuredResult) and #23 (deadline does not abort GET/sleep). This PR does not change the wait predicate or timeout bounding.

Evidence

Vitest, mocked fetch, no network, no phone call.

Before (published 0.7.0 probe):

[createAndWait: POST ok, GET rejects]  threw TypeError: fetch failed   callId missing
[createAndWait: POST ok, GET 502]      threw CalleAPIError             callId missing
[fetch rejects on POST create]         threw TypeError: fetch failed

After:

pnpm test
 ✓ tests/calls.test.ts (10 tests)

 Test Files  5 passed (5)
      Tests  39 passed (39)

pnpm run typecheck
# exit 0

New cases: rejected fetch on create is CalleConnectionError; POST-ok then GET reject or GET 502 includes callId: "call_123".

Test

pnpm test and pnpm run typecheck as above.

Source

From the CALL-E hackathon feedback lab: https://github.com/Arshgill01/ExactRef/blob/main/docs/feedback/lab/XR-605-create-and-wait-loses-call-id.md

Made with Cursor