perf: skip unused Safe Inventory pass in managed manifests

[dayebishouji]

In a managed-scope repository, businesssource.Build built a Safe Inventory and then called managedstate.Load, which built another one. The first inventory was never read by the managed branch. Build the standalone inventory only for the legacy branch; managed manifests still use the validated inventory and fingerprints from managed state. The final Overview source guard is unchanged.

Refs #127. This is a small reduction in server-side work, not a claim to resolve the full first-session latency reported in the issue.

Validation:

• The focused business-source and governance package tests passed before and after the change.
• On the same aligned 512-Entry fixture, the complete source manifest --json output before and after was byte-for-byte identical (409,172 bytes).
• Fresh MCP sessions in ABBA order (before, after, after, before) returned the same index SHA, six chunks, and 219,413 response bytes. Median first-Overview time was 3.656 s before and 2.604 s after; median time for all six chunks was 20.471 s before and 16.022 s after. These are two runs per build on a warm Windows filesystem cache, so they measure this fixture rather than a general speedup guarantee.
• AOCI verify and check passed with no drift; the Code Entry and Baseline were updated.
• go vet ./... passed. go test ./... -count=1 passed all packages except two Windows failures (internal/cli's initial Guide text assertion and internal/codebatch's unwritable-directory assertion); both exact tests fail the same way on an untouched origin/main checkout. The focused businesssource and volumegovernance packages passed.