docs: state Wagner's equal-weights placement contract; pass min_steps

[ms609]

Fixes #5
Fixes #6

#5 (T-369)

AdditionTree()'s numeric concavity never affected the returned topology — placement always uses an equal-weights Fitch proxy (fitch_indirect_length_cached), and concavity/min_steps only enter the final score_tree() call. That contract was already documented in a prior commit on @param concavity; this PR:

• Fixes the remaining gap: searchArgs never passed min_steps, so a finite concavity scored the internal result$score against min_steps = 0 instead of the dataset's real per-character minimum (MinimumLength(dataset, compress = TRUE), matching the idiom already used in MaximizeParsimony()/Resample()/SuccessiveApproximations()). AdditionTree() discards result$score, so this has no user-visible effect — corrected purely for internal correctness and to stop future confusion.
• Corrects a roxygen inaccuracy an external review caught: the doc claimed numeric concavity "affects only the reported $score", but AdditionTree() never returns a score at all. Reworded to say numeric concavity has no effect on AdditionTree()'s output, full stop.
• Adds regression tests: a mock-capture test pinning that AdditionTree() forwards real min_steps to ts_wagner_tree (fails pre-fix, confirmed); a direct ts_wagner_tree comparison showing min_steps changes the score but never the edge set; a characterisation test locking in that numeric concavity never moves AdditionTree()'s topology; and two small guards (that concavity = Inf still forwards min_steps = integer(0), and that concavity = "profile" still runs cleanly now that min_steps is computed after profile recoding).

Explicitly left undone (per the issue's own framing, a separate enhancement): using the weighted scorer for placement instead of the EW proxy. That would change start trees package-wide and needs its own wall-clock/reach A/B — out of scope here.

#6 (T-371)

Comment-only. Adds a note on wagner_goloboff_scores()'s tip_ambiguous computation (src/ts_wagner.cpp) recording that it ANDs across all n_states including the NA/inapplicable plane, so a tip coded {0,1} (every applicable state but not -) is scored non-ambiguous/informative despite carrying no information about which applicable state holds. wagner_entropy_scores()'s weight-blindness comment matching Goloboff's was already present on cpp-search from a prior pass, so nothing more was needed there.

Explicitly left undone: changing either behaviour (weight-awareness of the entropy scorer, or treating {0,1} as ambiguous). Both are judgement calls that shift start trees and need their own A/B — this PR only makes the existing behaviour legible.

Validation

• Build via the tarball recipe (never in-place); .o/.dll cleaned after every roxygen2::roxygenise()/check_man() pass per the debug-contamination gotcha.
• tests/testthat/test-AdditionTree.R: 36/36 pass against a temp-library install (not load_all()).
• devtools::check_man(): clean.
• spelling::spell_check_package() and the exact tests/spelling.R invocation: clean.
• Reviewed by an external-reviewer agent pass, scoped to correctness of the documented contract and whether min_steps genuinely leaves the topology untouched (verified structurally in src/ts_wagner.cpp and empirically across all 30 inapplicable.phyData datasets: 60/60 topologies identical). Two findings from that review (the $score doc inaccuracy, and a mislabelled/soft test assertion) are fixed in the second commit.
• GHA (agent-check.yml) is currently failing on this branch, but for an unrelated, already-tracked reason: pak cannot resolve the MaxMin Suggests dependency (agent-check.yml windows job fails on every branch: MaxMin is not published in ms609.github.io/packages, so pak cannot resolve it #43, in-progress), reproduced identically on cpp-search itself at the time of dispatch (run 30929069059, predating this PR's dispatch). Not caused by this change.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com