chore(coverage): close every reachable coverage gap in lib

[yordis]

• Line coverage across lib/ was 96.3%, and every uncovered line was a branch a caller can reach. An untested branch is a branch a regression can break silently, so the suite was not actually guarding them.
• The repository total is now 99.1%. Every remaining gap is a clause that cannot run against the dependency versions pinned here: gun 1.x compatibility fallbacks that the "Test (Gun 1)" job does exercise, a Finch pre-0.22 error shape, hackney 4.x, and a {:data_error, _} that OTP 24 through 28 never raise.
• Four clauses turned out to be unreachable rather than untested, so they were removed instead of being papered over with a test that could not exist:
  • sem_conv.ex had an extract_port/1 clause for scheme: "https" with a non-integer port. URI.parse/1 always fills the scheme default port, verified across eleven URL shapes including https://host:/p, https://host:abc/p and https://host:-1/p, all of which yield 443.
  • test.ex had a catch-all encode!/2. Its only call site passes the literal "application/json", and the two application/json clauses are total.
  • mock.ex had a raise other fallback in agent_set/1. start_supervised/2 returns only {:ok, pid} or {:error, {:already_started, pid}} there, so the clause was both dead and latently broken: raise on a tuple raises ArgumentError about raise/1 rather than the intended error.
  • The httpc adapter had a fallback for a multipart request without a content-type header. Multipart.headers/1 always returns one, so List.keytake/3 can never miss.
• Every new test was checked by mutation testing rather than by coverage alone. Coverage proves a line ran; it does not prove an assertion would notice if the line changed. Fifty two mutants were applied across the touched modules and all fifty two were caught, with one exception that was triaged as equivalent: deleting the max_retries: 0 shortcut in retry.ex changes nothing observable, because put_retry_count_opt/2 is already a no-op at retries: 0.
• Two of those mutants only died after the tests were rewritten, which is the point of running them: the httpc and ibrowse binary body tests passed while the adapter appended a trailing space, and the gun connection reuse tests passed while the scheme comparison was deleted outright.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[cursor]

PR Summary

Low Risk
Changes are almost entirely tests plus removal of unreachable library branches; runtime behavior for supported call paths should be unchanged.

Overview
This PR raises lib/ test coverage by exercising reachable branches across adapters, middleware, mocks, and core Tesla behavior, and deletes three clauses that could never run instead of adding fake tests for them.

Production tweaks (dead code only): Tesla.Mock no longer has a broken raise other fallback when the global mock Agent fails to start; OpenTelemetry extract_port/1 drops a redundant https → 443 clause because URI.parse/1 already supplies an integer port; Tesla.Test.encode!/2 loses an unused catch-all since callers always pass "application/json".

New / expanded tests guard real edge paths: Finch (invalid :response, non-HTTP responses, chunked trailers) and Gun (reused conn on wrong port/scheme, stream errors to multiple owners, monitor :killed, retry while connections flap) via raw TCP helpers; httpc/ibrowse binary body passthrough; middleware error passthrough and encoding options (JSON/MessagePack streams and custom encode/decode, Logger OTel with charlist URLs and :default log level, Query empty-map merge rules, Retry with max_retries: 0, SSE id/retry, PathParams missing values); Mock global replacement and registered $ancestors; plus smaller cases for multipart header validation, Tesla.effective_adapter/1 app config, builder deprecation stderr, and anonymous function middleware in Tesla.run/2.

^{Reviewed by Cursor Bugbot for commit 30d0fe9. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit a489671. Configure here.}