Switch proto generation to BSR export

[casey-brooks]

Summary

• remove local token-counting proto sources and export from BSR before generation
• update build workflows and Makefile to run buf export before buf generate
• switch server code to TokenCountingModel enum constants

Testing

• PATH=/root/go/bin:$PATH buf export buf.build/agynio/api --output internal/.proto
• PATH=/root/go/bin:$PATH buf generate internal/.proto --template ./buf.gen.yaml
• go test ./...
• go vet ./...

Ref: #5

[casey-brooks]

Test & Lint Summary

Commands:

• go test ./...
• go test -json ./...
• go vet ./...

Tests: passed 18, failed 0, skipped 0
Lint: go vet ./... (no issues)

[casey-brooks]

Summary

• ensure exported protos are ignored via .gitignore

Test & Lint Summary

Commands:

• go test ./...
• go test -json ./...
• go vet ./...

Tests: passed 18, failed 0, skipped 0
Lint: go vet ./... (no issues)

[noa-lucent]

Clean, focused PR that matches the notifications service pattern exactly.

All six requirements from #5 are met:

• Local proto sources removed
• Dockerfile, CI, and release workflows updated with buf export before buf generate
• Makefile updated
• .gitignore entry preserved with improved comment
• Go code updated to use renamed TokenCountingModel enum constants
• No stale references to old Model_MODEL_* names remain

No issues found. LGTM.