fix: Check the error from constructing the event poll request

[keelerm84]

Summary

eventLoop discarded the error from http.NewRequest and set a header on the
returned request immediately. On a construction failure that request is nil, so the
daemon panicked on a nil pointer dereference instead of reporting the problem.
featureLoop already checks the identical call, so the two loops now agree.

The loop returns the error rather than logging it and polling again. pollURI is
computed once from SALESFORCE_URL before the loop starts, so a URL that
url.Parse rejects fails identically on every iteration -- retrying would spin
forever with no way to make progress, and would bury an operator misconfiguration
in a log line.

The new test drives eventLoop with a SALESFORCE_URL holding a control
character, which is what makes url.Parse fail. It recovers a panic inside the
loop goroutine, so the previous behavior surfaces as one test failure rather than
killing the test binary. Verified to bite: with the check removed the test fails
with invalid memory address or nil pointer dereference.

---

Note

Overview
eventLoop now checks the error from http.NewRequest before setting headers on the poll request, matching featureLoop. A bad SALESFORCE_URL (e.g. one url.Parse rejects) used to yield a nil request and crash the daemon on header setup; the loop now returns that error instead of retrying, because pollURI is fixed for the process lifetime so retries would only spin.

A new test drives eventLoop with an invalid Salesforce base URL, asserts no panic, expects the same construction error as http.NewRequest, and times out if the loop keeps polling.

^{Reviewed by Cursor Bugbot for commit 227ea39. Bugbot is set up for automated code reviews on this repo. Configure here.}