Add BenchmarkFilter_15Items_WithFullContent, BenchmarkFilter_15Items_IncludeFullContent,
and BenchmarkFilter_LargeDataset_WithFullContent. Items are seeded with realistic
~2KB full_content. Results show excluding full_content reduces B/op by ~40%
(363KB vs 595KB for 15 items).

https://claude.ai/code/session_01DpWhB9uGGMBnzqS28HxnuV

- CSRF: test excluded paths (/api/login, /api/logout), PUT/DELETE methods
- Item model: test Filter includeContent flag, ItemById returns content,
  multiple feed_ids filtering
- API: test read_filter=all param, feed_ids comma-separated filter,
  full_content exclusion from stream
- Routing: add v3 frontend route test

https://claude.ai/code/session_019Z4VJxzY7tcAuNkPAkvry9

The /stream endpoint now skips selecting full_content and header_image
from the database, reducing memory allocation and response payload size
for list views. Full content remains available via /item/:id endpoint.

Adds omitempty to JSON tags so empty fields are omitted from responses,
and a test verifying full_content is excluded from stream responses.

https://claude.ai/code/session_019Z4VJxzY7tcAuNkPAkvry9

Go benchmarks cover item CRUD/filter/sanitization, API endpoints (stream,
item update, feed list), middleware stack (gzip, security headers, CSRF),
and crawler pipeline (feed parsing, mocked crawl). Stress tests verify
concurrent reads/writes and large dataset handling. Frontend perf tests
measure template generation, DOM insertion, and store event throughput.

New Makefile targets: bench, bench-short, stress, test-perf.

https://claude.ai/code/session_01ChDVWFDrQoFjMYHpaLGr9s