docs: add technical reports for #333 through #339 (#340)

## Summary

Adds bilingual (English/Korean) technical reports for the seven PRs merged after PR #331's batch, following the `TECHNICAL_REPORTS/` convention established there: `<PR-number>-<kebab-slug>-<YYYYMMDD>.{en,ko}.md`, numbered sections, tables, and an explicit verified-versus-assumed distinction.

| PR | Issue | Subject |
|---|---|---|
| #333 | #324 | metrics baseline (`all_smi_up`, build-info) plus the `/-/ready` endpoint |
| #334 | #326 | TUI survives a terminal reporting no size |
| #335 | #329 | both service smoke tests gated on `/-/ready` |
| #336 | #330 | both systemd scopes run unconditionally, with teardown |
| #337 | #325 | Apple Silicon GPU metrics omitted instead of reported as 0 |
| #338 | #327 | SSE test measures publish cost, not sleep overshoot |
| #339 | #328 | Docker image built on pull requests |

Each report was written after reading the PR body, the full diff, and the originating issue, and claims were verified against the diff rather than copied from the PR description.

## Files

14 files, all under `TECHNICAL_REPORTS/`:

- `333-metrics-baseline-ready-endpoint-20260806.{en,ko}.md`
- `334-tui-zero-size-terminal-fix-20260806.{en,ko}.md`
- `335-service-smoke-tests-ready-gate-20260806.{en,ko}.md`
- `336-systemd-scopes-unconditional-teardown-20260806.{en,ko}.md`
- `337-apple-silicon-gpu-metric-omission-20260806.{en,ko}.md`
- `338-sse-publish-cost-test-fix-20260806.{en,ko}.md`
- `339-docker-build-pull-requests-20260806.{en,ko}.md`

## Notes for the reviewer

- PR #334's report flags one claim it could not confirm from the diff: the PR body states `cargo clippy --bin all-smi --tests` caught a `pub` item live in the library target and dead in the binary target, but the merged diff shows no trace of what was fixed (no `pub fn new` or similar was removed from `ui/viewport.rs`). Recorded as unverified in the report rather than asserted as fact.
- PR #339's report draws the `linux/arm64`/`FindBin` finding from the linked issue's comment thread rather than the PR body or diff, since it does not appear in either.

## Test plan

- [x] Each report's factual claims were checked against `gh pr diff` for the corresponding PR, not only the PR description.
- [x] Proofread the Korean reports for translationese and transliteration errors.
- [x] Verified no em dashes and no AI-attribution text anywhere in the new files.