fix: correct bgl entrypoint command guard

[tolga-tom-nook]

Summary

• correct the Docker entrypoint command guard so option and .conf invocations are rewritten to bgl_oneshot instead of the non-existent bgl command
• add a focused entrypoint regression test covering option and config-file startup forms

Verification

• bash test/tests/entrypoint.sh
• git diff --check origin/main..HEAD

Bounty

Submitted for Bitgesell improvement bounty consideration: BitgesellOfficial/bitgesell#81 and BitgesellOfficial/bitgesell#39.
Payout address, if approved: 0x4a76c7E64C08cF29B59eFC640b4ada97A270d428

---

Executor revalidation (20260618T184217Z)

Fresh payout-readiness check after the review-assist note on this PR:

• Live PR state: open, authored by tolga-tom-nook, mergeable/clean, not draft.
• Head: bdf07c876f48455b6c09a05fcb3c005572a9fe8f.
• Base sync: git rev-list --left-right --count origin/main...HEAD -> 0 2.
• Validation: bash test/tests/entrypoint.sh passed.
• Validation: git diff --check origin/main...HEAD passed.

This remains a focused entrypoint/default-command guard fix for the Bitgesell bounty lane; no runtime scope expansion or payout change.

[tolga-tom-nook]

Pushed a focused test-hardening follow-up (bdf07c8) for the entrypoint regression.

Added coverage that explicit commands still pass through unchanged while -flag and *.conf invocations are rewritten to bgl_oneshot. This tightens the proof that the fix only restores the intended default-command behavior and does not hijack custom container commands.

Validation rerun locally:

bash test/tests/entrypoint.sh
git diff --check

[MyTH-zyxeon]

Maintainer-facing review-assist for the Bitgesell #81/#39 bounty lane:

I checked this as a BGL ARM Docker entrypoint/runtime fix. The useful part is that it removes the stale btc_oneshot guard and makes the default-command rewrite use bgl_oneshot, while adding regression coverage for the two default startup forms and custom-command passthrough.

Acceptance checks I would use before landing:

1. bash test/tests/entrypoint.sh should pass from the repo root and prove -flag and *.conf invocations rewrite to bgl_oneshot.
2. The explicit command passthrough case should remain unchanged, so container users can still run custom commands without being redirected into the node entrypoint.
3. The added command -v gosu and id BGL checks should be acceptable as fail-open behavior when a derived image lacks gosu or the BGL user; if maintainers expect hard failure instead, that policy should be stated before merge.
4. The test path is test/tests/entrypoint.sh; if the repo has an intended aggregate test runner, wiring this script into it would make the regression harder to miss.

I did not run local Docker or shell tests in this triage pass; this note is based on the visible diff and PR metadata only.