Fix get_org_id on SaaS + release 0.4.1

[Doug Guthrie (dpguthrie)]

Problem

A user on 0.4.0 hit this migrating logs into SaaS:

Failed to migrate logs: BraintrustAPIError: Unable to determine org_id for
attachment operations. Tried /ping and /v1/ping; last error: 404 Not Found
for url 'https://api.braintrust.dev/v1/ping'

get_org_id() only tried GET /ping and /v1/ping, which 404 on the SaaS REST API (api.braintrust.dev). The 0.4.0 oversize-field attachment spilling calls get_org_id() for every row over the ~20 MB per-span limit, so any real migration with a large span fails. (The SDK resolves org_id via POST {app_url}/api/apikey/login on the app URL, which this tool doesn't use — it only has the API URL.)

Fix

Resolve org_id from a project instead. org_id is a documented top-level field on every Project object (confirmed on both the list and get-by-id responses), and every project visible to an org-scoped API key is in that org:

• Capture opportunistically whenever the client lists or creates a project (_maybe_capture_org_id). The orchestrator discovers/creates the destination project before streaming logs, so in the normal flow org_id is already known by the time spilling happens — no extra request.
• Fallback: get_org_id queries GET /v1/project?limit=1 when nothing has been captured yet.
• Dropped the /ping probes entirely — they only 404 on SaaS and add nothing now.

Also repairs the latent copy_attachments path, which had the same get_org_id dependency.

Why it holds for the live case: by the time logs are migrated (and spilling), the destination project already exists (created earlier in the run), so its org_id has already been captured.

Tests

• test_get_org_id_captured_from_list_projects / _from_create_project — captured with zero extra requests.
• test_get_org_id_queries_project_when_not_cached — fallback query.
• test_get_org_id_raises_when_no_project — clear error.
• Updated the attachment-copier and oversize-spill mocks to serve /v1/project.
• Full unit suite: 292 passing.

Release

Targets a 0.4.1 patch — blocking a live migration. Self-contained off main (one commit, client.py + tests); independent of the refactor PR #20.

🤖 Generated with Claude Code

[Doug Guthrie (dpguthrie)]

Added the 0.4.1 release prep to this PR: CHANGELOG entry, version bump (pyproject.toml, __version__, uv.lock), and README pin → v0.4.1. Merging this leaves main immediately taggable as v0.4.1.

Verified the fix end-to-end against live SaaS (seeded a 1.5 MB span, lowered the spill threshold to 1 MB, migrated): the field spilled to a braintrust_attachment, the dest span carried the reference, and the downloaded attachment matched the original bytes (content_matches_original=True). The POST /attachment (attachment_metadata_dest) call that previously 404'd now succeeds.