chore(release): cut 0.1.37

[Jason Robert (jrob5756)]

Release 0.1.37

Previous tag: v0.1.36 (published 2026-09-02) → new version 0.1.37, tag v0.1.37.

Commit audit

All 12 commits in v0.1.36..main were audited; every one has an explicit disposition.

#	Commit	PR	Classification	Changelog
1	bf35856	#494	Internal — CI action bump	Omitted (intentional)
2	ead6c76	#502 (issue #501)	Added	Already present
3	ff4b312	#503	Added + Changed	Relocated — see below
4	db1d3a4	#506	Fixed	Already present
5	544a2bf	#507	Fixed	Already present
6	ecb93a7	#515	Internal — uv.lock transitive bump	Omitted (intentional)
7	e5e5a75	#520	Internal — frontend devDependency	Omitted (intentional)
8	0939597	#516	Internal — uv.lock transitive bump	Omitted (intentional)
9	b2ad333	#514 (issue #513)	Added + Changed	Already present
10	b2f8871	#512	Fixed	Added — was missing
11	eb1b76a	#511	Fixed	Present; PR link added
12	3becc66	#510 (issue #509)	Fixed	Already present

Changelog corrections

Three defects were found and fixed while reconciling:

1. feat(providers): always-on client-side context compaction for pydantic-ai providers (claude, openai) #503's entries were filed into the already-published 0.1.36 section. That PR
   merged 2026-09-04, two days after v0.1.36 was published, but appended a second
   ### Added and ### Changed block to the 0.1.36 section instead of [Unreleased].
   The released notes therefore advertised always-on context compaction, the
   provider-advertised token-limit metadata, and the max_tokens 8192→16384 default
   change as shipping in v0.1.36 — none of which are in that tag. Verified against
   git show v0.1.36:CHANGELOG.md, which contains no mention of compaction. Those four
   entries now live in 0.1.37, and the 0.1.36 section is byte-identical to the
   published tag again.
2. fix(providers): prevent concurrent registry creation #512 was undocumented. Added a Fixed entry for the provider-registry race.
3. [Unreleased] had two separate ### Fixed headings. Consolidated; the release
   section is now ordered Added → Changed → Fixed per Keep a Changelog.

Resulting 0.1.37 section: 4 Added, 3 Changed, 5 Fixed.

Validation

Gate	Result
make check (ruff check, ruff format --check, ty)	passed
make validate-examples	passed
uv run pytest -m "not real_api and not performance and not install_scripts"	8463 passed, 32 skipped

The release-gate marker expression matches release.yml and ci.yml
(-m "not real_api and not performance"). A bare make test additionally runs
@pytest.mark.performance tests; test_backtick_wall_scales_near_linearly is
wall-clock sensitive and flakes under full-suite load — it passes in isolation
(9 passed in 0.36s) and is excluded from both CI and the release workflow.

Diff scope

Only the three release files: CHANGELOG.md, pyproject.toml, uv.lock. The
lockfile change is solely the conductor-cli project version (no dependency movement).

Next step

Merging this PR will be followed by tagging the merge commit v0.1.37 and pushing it,
which triggers .github/workflows/release.yml to build
and publish the GitHub Release with its artifacts.