Move the Erlang services off lager onto OTP logger

[tas50]

Replaces every lager call in oc_erchef, oc_bifrost and bookshelf with the OTP logger macros from kernel/include/logger.hrl, and drops the lager dependency, its parse transform and its application entries.

86 call sites across 31 files, plus 5 rebar.configs and 6 .app.src files.

Why

lager is unmaintained and reaches inside OTP in four places rather than merely using it:

• lager_stdlib.erl is a verbatim copy of unexported OTP stdlib functions — "Functions from Erlang OTP distribution that are really useful but aren't exported", Copyright Ericsson AB 1996-2009 — and lager_format.erl is a fork of io_lib_format (1996-2011-2012).
• lager_app starts the deprecated error_logger, installs OTP's internal error_logger module as a logger handler by hand rather than through error_logger:add_report_handler, then calls logger:remove_handler(default) to delete OTP's own default handler. Its own comment calls this "a band-aid" for OTP 21.
• error_logger_lager_h pattern-matches OTP's internal {error_report,_} / {info_report,_} shapes.
• lager_transform rewrites the AST at compile time — hence "lager has to come first" in every rebar.config.

Upstream's last commit is e6b3178 (2023-11-02), before OTP 27 existed, so none of that is being re-synced.

To be precise about scope: lager compiles fine on OTP 27 — I checked. This is not a compile-time blocker. The case is unmaintained code reaching into OTP internals, which is runtime fragility and accumulating risk, not a build break.

?LOG_* macros rather than logger:Level/2 calls: the macros capture module, function and line, which is the job lager_transform did. Plain function calls silently lose that metadata.

Per service

• oc_erchef — 74 sites across 25 files, all plain level calls (no lager:md, sinks or traces). lager:start/0 goes from a test helper; logger is part of kernel and always running.
• bookshelf — 10 sites. internal.hrl already defined ?LOG_INFO/?LOG_ERROR/?LOG_DEBUG at exactly the arities logger.hrl provides, so including it there migrates every existing ?LOG_* call site untouched and retires the deprecated error_logger:info_msg/error_msg it used for two of them.
• oc_bifrost — 2 sites.

Things only a build finds

These are not in the obvious places and are why this touches 31 files rather than 30:

• goldrush in oc_chef_wm.app.src — lager's own dependency. Leaving it behind breaks rebar3 dialyzer ("Could not find application: goldrush") and rebar3 release ("Application needed for release not found"), after every unit test has already passed. Removed.
• Sub-app rebar.configs — apps/data_collector and apps/chef_telemetry each declare lager and the parse transform independently of the top-level config, on a rolling master branch.
• Six oc_chef_wm itest suites carrying -compile([{parse_transform, lager_transform}]), setup_helper setting lager's error_logger_redirect and starting lager/goldrush, and the bookshelf SUITE calling lager_common_test_backend:bounce/1. None are built by the default profile.
• The relx release app list in oc_bifrost/rebar.config.
• Every -include_lib was checked to be outside -ifdef blocks — an include placed inside -ifdef(TEST) silently vanishes from normal builds.

Verified

erlang:26 containers, clean _build:

• oc_bifrost, bookshelf and oc_erchef all compile with warnings_as_errors on.
• oc_erchef assembles: Release successfully assembled: _build/default/rel/oc_erchef.
• oc_erchef dialyzer is clean — 160 files analysed, exit 0. This matters specifically: Resolving project files is the step that fails on a stale goldrush entry, so a clean run here is what confirms the .app.src cleanup is complete.

This does not remove lager from the build yet

Three libraries still declare lager and are pulled in transitively, so it remains in _build and in the assembled release until they land:

Library	PR	Sites
chef_secrets	chef/chef_secrets#88	7 + its applications entry, which is what starts lager
folsom_graphite	chef/folsom_graphite#16	14
opscoderl_httpc	chef/opscoderl_httpc#21	1

Merge those three first. The lockfiles are deliberately left alone here — re-pinning them now would mean pointing at fork branches.

The sys.config templates that carry the {lager, [...]} blocks live in chef-server-omnibus-config, not this repo, so converting those to logger handlers (and moving lager's $D0 midnight rotation to logrotate, since logger_std_h has no time trigger) is a separate change there.

CI note

buildkite and the Grype scan fail on this PR. They fail identically on every open PR in this repo, including #4239, #4230 and #4209 — Grype reports 1 CRITICAL, 3 HIGH from the existing dependency tree, gated by grype-fail-on-high: true. Neither is related to this change.