Split title/description/details and use base-R macros for package Rd

[TroyHernandez]

Summary

• Parser now splits pre-tag content into title (para 1), description (para 2), details (para 3+) on blank lines, matching roxygen2. Previously paragraphs 2+ were jammed into description and \details{} never got populated.
• generate_package_rd() uses base-R Rd macros (\packageTitle{}, \packageDescription{}, \packageAuthor{}, \packageMaintainer{}, \packageIndices{}) instead of baking content into the Rd file. DESCRIPTION is the source of truth.
• Hand-written paragraphs after a _PACKAGE block's description now land in \details{} — the slot for design notes, limitations, etc.
• Drop a now-dead get_maintainer_from_desc() helper.

Triggered by Dirk Eddelbuettel's review of tinyrox 0.3.3 against rspdlite.

Test plan

• 145 tests pass (was 119; added 26 covering the new parser states and macro template)
• tinyrox::document(); tinypkgr::install(); tinytest::test_package('tinyrox') clean
• Against eddelbuettel/rspdlite: man/formatter.Rd is byte-identical to roxygen2's output except for the header comment; man/rspdlite-package.Rd correctly populates \details{} (was missing) and uses macros for title/description/author
• tools::checkRd() on the regenerated rspdlite-package.Rd returns clean
• Verified on corteza (an unrelated downstream): no empty \details{} blocks after fresh install

Visible behavior change

For packages with a _PACKAGE block, the generated pkg-package.Rd switches from hardcoded title/description/author to macro-based delegation to DESCRIPTION. Same content if R-file and DESCRIPTION agree; different if they diverge. Also adds a "Package Content" function index (was missing).

Out of scope (separate follow-ups)

• ##' (legacy two-hash) doc prefix isn't recognized — R/parse.R:17 only matches ^#'. Surfaced because rspdlite's R/wrap.R uses this style.
• Dirk reported a residual issue with namespace = "append" mode that wasn't reproducible against rspdlite (it has no @export/@import tags). Needs a different repro.

[eddelbuettel]

Dirk reported a residual issue with namespace = "append" mode that wasn't reproducible against rspdlite (it has no @export/@import tags). Needs a different repro.

I was sufficiently unclear, my bad. I ran my littler script trox.r, it just called document(). This nuked my existing (manual) NAMESPACE (but hey git saves our bacon,...). Running with document(namespace="none") as I should have in the first place did the job more nicely.

[TroyHernandez]

There's a real UX question buried in this: default behavior of document() silently nuking a hand-edited NAMESPACE if you forget the flag. Possible follow-ups for another PR:

• Detect-and-warn: if NAMESPACE exists and has no tinyrox header line, warn before overwriting.
• Change default to "append": less destructive default. Existing tinyrox start/end markers protect the user's hand edits.

Neither is in scope for this PR. Just worth flagging.

[eddelbuettel]

Agree completely--it's an open question without a perfect answer. Maybe even a defensive posture is good: do not overwrite a NAMESPACE unless it as the tinyrox tag in it?