ci(apm): preload PoC basic testing improvements

[DavSanchez]

Summary

This PR builds on top of #2517 and proposes a few improvements to the preload agent E2E scenario, while keeping the changes scoped to a PoC for APM support on-host through the preload library.

Why not commit the agent type definition

The original branch added agent-control/agent-type-registry/newrelic/com.newrelic.preload-0.1.0.yaml to the main codebase. Since this work is still a PoC, we shouldn't be shipping a com.newrelic.preload agent type as part of the registry yet. Instead, this PR:

• Removes the YAML from agent-type-registry/.
• Inlines a minimal version of the agent type definition inside the E2E scenario, written at runtime to the dynamic agent types directory (/etc/newrelic-agent-control/dynamic-agent-types/preload.yaml).

This way, the PoC scenario can exercise the full flow (Agent Control loading a custom agent type, downloading the OCI artifact, etc.) without leaking PoC artifacts into the distributed registry.

Test scenario improvements

A few minor assertions were added to the PreloadAgent scenario so that failures are caught early and surface clear errors instead of failing later in opaque ways while the actual test scenario takes shape:

• Wait (with retries) until the preload OCI package is downloaded and extracted under the AC-managed stored_packages/preload-agent directory.
• Locate the shared library (.so) inside the extracted package and fail loudly if none is found.
• Append the resolved .so path to /etc/ld.so.preload (previously the path was a placeholder and the file path itself was wrong: /ec/ld.so.preload).

The local agent config sent to AC was also trimmed to the minimum needed for this PoC (just version), since the previous fields were not consumed by the agent type used here.

CI changes

The component_onhost_e2e.yaml workflow was updated so the new scenario actually runs:

• Added preload-agent to the scenarios matrix.
• Introduced HOST_E2E_PRELOAD_VERSION and threaded it through as --preload-version to the e2e runner.

[mvicknr]

Claude found this issue:

🔴 Bug: ${{ ... }} won't be substituted by Agent Control

test/e2e-runner/src/linux/scenarios/preload_agent.rs:50-77 — the YAML is a plain r#"..."# raw string, not a format! template, so the doubled braces are not escapes — they end up literally in the file:

repository: ${{nr-var:oci.repository}}
version: ${{nr-var:version}}

Compare with agent-control/tests/on_host/scenarios/oci_auth.rs:43-54, where ${{...}} is inside format!(...) and correctly resolves to ${...}. AC's template engine expects the single-brace form
(${nr-var:version}), so the inlined agent type will fail to resolve version and the OCI download won't get a valid tag.

Fix: use single braces (${nr-var:...}) since this is a raw string, or wrap the literal in format! if you want to keep the doubled form for visual consistency with the other tests.

[DavSanchez]

Fixed e91cd32

[mvicknr]

Also from Claude;

🔴 Bug: so_path extraction picks up the wrapper text, not the actual path

test/e2e-runner/src/linux/scenarios/preload_agent.rs:115-122:

let so_path = exec_bash_command(&find_so_command)...;
let so_path = so_path.lines().last().unwrap_or("").trim().to_string();

exec_bash_command (test/e2e-runner/src/linux/bash.rs:22) does not return raw stdout — it returns a wrapped, multi-line string:
command

success
Stdout:
Stderr:

The Stderr: line is always last, so so_path ends up as "Stderr:" (or empty after the colon depending on trim). The if so_path.is_empty() guard does not catch this, and the next step appends Stderr: to
/etc/ld.so.preload. None of the existing callers parse stdout from exec_bash_command — this is the first one, and it doesn't work.

Fix options (pick one):

• Add a sibling helper that returns just stdout (e.g., exec_bash_command_stdout), then use it here.
• Run find ... -print -quit on the host and have the helper write its result to a file you read_to_string.
• Parse the wrapper: extract the line starting with Stdout: and strip the prefix — least preferred since it bakes in a fragile dependency on the wrapper format.

[DavSanchez]

Changed the strategy used here: fec0466