cd_trend(): typed zero-row table when no series is long enough

[NewGraphEnvironment]

Summary

• cd_trend() now returns a zero-row tibble with its full column set when no combination has 3+ years, instead of a shapeless 0 x 0. cd_summary() on it returns an empty table (it used to fail with Column 'period' not found), and cd_plot_timeseries() no longer warns about uninitialised columns.
• The template's key types come from the same expand.grid the rows are built from, so a factor variable/period stays a factor and trend_start = NULL keeps its column. slope/intercept lose zyp's "yr"/"Intercept" names, so empty and full results share one ptype.
• @return in ?cd_trend documents the empty shape; CLAUDE.md notes it on the trend-table contract.

Fixes #101

Related Issues

• Relates to NewGraphEnvironment/sred-2025-2026#23

Test plan

• devtools::test(): FAIL 0 | PASS 436 (the 6 warnings are pre-existing in test-cd_plot_comparison.R)
• Seven new tests plus one extended, all failing on main; each type guard fails against its own restored defect (mutation run in a scratch copy)
• devtools::check(): 0 errors. The 1 warning and 8 notes are pre-existing (no VignetteBuilder, .Rbuildignore gaps tracked in .Rbuildignore: planning/, CLAUDE.md, scripts/, data-raw/ ship in the package tarball #100 / .Rbuildignore misses planning/, scripts/, data-raw/, logs/, dev/, .claude/, CLAUDE.md - all ship in the tarball #107, .data bindings, unused sf)
• /code-check: 3 rounds, plus a plan review. Round 1 found the factor and NULL regressions and they were fixed; rounds 2 and 3 were clean, and round 3 checked all 11 template columns against the row builder

Notes

• Behaviour change: names(trn$slope) / names(trn$intercept) are now NULL. Nothing in the repo reads them.
• The existing short-series test asserted only nrow == 0, which the 0 x 0 already satisfied. That is why this went uncaught.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GignjoKpZADXywTf6pAMgj