fix(github): create domain accounts for non-committer authors (#8886)

[JAORMX]

Problem

issues.creator_id and pull_requests.author_id are written for every author, but a domain accounts row is only created for users we collected a full profile for (effectively, committers). Authors who only filed an issue, or opened a PR without committing, become orphan foreign keys:

SELECT i.url, i.creator_id, i.creator_name, a.user_name
  FROM issues i
  LEFT JOIN accounts a ON a.id = i.creator_id
 WHERE i.url = 'https://github.com/stacklok/toolhive/issues/4297';

creator_id is set and creator_name is milichev, but a.user_name is NULL. Bot filters keying on accounts.user_name LIKE '%[bot]' miss those rows for the same reason.

Fixes #8886.

Root cause

The issue and PR extractors already record every author in _tool_github_repo_accounts. But ConvertAccounts sourced the domain accounts table FROM _tool_github_accounts, which is only populated for users we fetched full profiles for. So the convertors generate creator_id / author_id for everyone, while accounts only ever gets the committers.

Change

• ConvertAccounts now reads FROM _tool_github_repo_accounts LEFT JOIN _tool_github_accounts, so every user the repo references becomes a domain account: enriched with name/email/avatar when we have the detail, login-only otherwise. The domain id uses the same didgen generator the issue/PR convertors use, so the foreign keys line up.
• pr_extractor also emits a repo_account for the merged_by user, which fixes pull_requests.merged_by_id (that user wasn't recorded anywhere before).
• The query is MySQL/PostgreSQL-agnostic (COALESCE not IFNULL, no backtick quoting, values parameterized via the dal). The join mirrors the existing one in account_org_collector.go.

Testing

• The e2e fixture didn't exercise the bug (every referenced account already had a profile row), so I added the orphan case from the issue: milichev, referenced by the repo with no detail row. It now gets a login-only accounts row.
• Added a referential-integrity assertion to TestAccountDataFlow: every account the repo references must resolve to a domain accounts row, generated with the same id generator the convertors use. Verified it fails against the old converter and passes with the fix.
• Full plugins/github/e2e suite passes on both MySQL and PostgreSQL.

Out of scope

Plural issue/PR assignees and PR requested reviewers aren't seeded into _tool_github_repo_accounts, so those FKs can still be unresolved. The github_graphql plugin likely shares the root cause. Happy to follow up on these separately.

🤖 Generated with Claude Code

[klesh]

_raw_data fields are missing, please fix it. Thanks.

[JAORMX]

Oops, will do when I'm back on my computer

[JAORMX]

Done in b67aad7! The converter now carries the _raw_data fields over: it prefers the enriched _tool_github_accounts row when we collected a profile, and falls back to the _tool_github_repo_accounts row for the non-committer case (that's why milichev shows _raw_github_api_issues provenance). The e2e fixture was missing the _raw_data columns too, so I added those and regenerated the snapshot... the enriched rows now carry exactly the same provenance as before this PR.

While at it, I also guarded pr_convertor.go against generating account ids for a zero AuthorId/MergedById. An unmerged PR would otherwise get merged_by_id pointing at github:GithubAccount:1:0, which is the same orphan shape this PR is fixing (issue_convertor already had that guard).

Verified the full github e2e suite on both MySQL and PostgreSQL.