fix(workbuddy): preserve cache usage from aggregated responses

[gruffwy]

Problem

WorkBuddy's non-streaming path is produced by aggregating an upstream stream. The aggregate conversion currently retains prompt/completion usage but drops cache-read and cache-write token counts.

Behavior

• Parse usage.cache_read_tokens and usage.cache_write_tokens from aggregated responses.
• Fall back to usage.prompt_tokens_details.cached_tokens when the top-level cache-read field is absent.
• Preserve explicit zero values by using pointer fields.
• Prefer the top-level cache-read value when both representations are present.

Tests

• go test ./internal/providers/workbuddy
• GOFLAGS=-buildvcs=false go test ./internal/app -run 'TestImportConstraints|TestDutyBoundaries' -count=1
• git diff --check