docs(rust): a category index on the crate front page, from the grouping the registry already had (#179 D6)

[kevinlincg]

D6: "1020 methods on one Core with no category index, prelude, or grouping. abstract_api::Group already encodes the grouping at runtime but is not surfaced as docs. 'How do I find the candlestick ones?' works only by alphabetical accident."

The alphabetical accident is real but partial. The 61 recognizers do all start with CDL, and nothing else does — so that one question happens to have an answer. It buys nothing for the other nine groups: AD, ADOSC, CMF, EFI, MARKETFI, NVI, OBV, PVI, PVO and VWAP are the volume studies, and their names say so only to a reader who already knows.

What this adds

rust_doc::category_index renders the grouping as //! lines that the lib.rs scaffolding substitutes into the crate docs: one ## heading per group with its member count, and under it one line per function — a link to that function's own page, then that row's hint.

//! ## Volume Indicators (10)
//!
//! * [`AD`](Core::AD) — Chaikin A/D Line
//! * [`ADOSC`](Core::ADOSC) — Chaikin A/D Oscillator
...

Groups come out in the order Group::ALL declares them: a BTreeMap over the display strings, which is the same order, so the front page and the registry enumerate categories identically.

Two fields, both straight off the definition — the name that is also the method's name, and the hint that becomes FuncInfo::hint and the FuncId variant's own doc line. No third source, so the index cannot say something about a function that the registry does not. That includes the empty-hint case: the registry stores "", and the index writes the link alone rather than falling back to description and reading differently from every other surface. No shipped function has an empty hint today; the rule is there so that adding one cannot make the two disagree.

What gates it, and what did not

rustdoc, already. [SMA](Core::SMA) naming a method that does not exist is rustdoc::broken_intra_doc_links, and the nightly's clippy job runs cargo doc under RUSTDOCFLAGS=-D warnings. Control, watched: rewriting one entry to Core::SMAX fails that step —

error: unresolved link to `Core::SMAX`
   --> library/src/lib.rs:179:15
    | //! * [`SMA`](Core::SMAX) — Simple Moving Average
    |               ^^^^^^^^^^ the struct `Core` has no field or associated item named `SMAX`
    = note: `-D rustdoc::broken-intra-doc-links` implied by `-D warnings`

Nothing gated an omission, and that is why the builder lives in rust_doc rather than inline in main.rs: there it is reachable from the test suite. A filter that quietly drops a function leaves a page that still builds clean and simply never mentions it. rust_category_index_lists_every_function_once walks the corpus and requires each name exactly once, no entry beyond the corpus, and each heading's count equal to the bullets under it. Both halves watched red:

sabotage	result
builder skips Pattern Recognition	CDL2CROWS must appear exactly once in the category index
heading count members.len() + 1	Cycle Indicators says 6 but lists 5

The cost

The front page grows: 108,739 → 137,217 bytes of rendered HTML (+26%), 215 lines of //! in a generated file. That is the whole of it, and it is the maintainer's call whether a longer front page is worth a category index; nothing is special-cased to hide it.

Doc build time is not part of the cost, and I am not claiming it is cheaper either: three warm cargo doc --no-deps runs each way gave 2.88 / 3.04 / 2.91 s before and 2.82 / 2.90 / 2.91 s after. The ranges overlap — the numbers do not discriminate.

Deliberately not done here

Each is a decision rather than a rendering, so each is yours:

• The prelude D6 also asks for. A public re-export set is API surface, not docs, and it freezes before the first publish.
• The same index in README.md. Core::SMA links resolve in rustdoc and nowhere else, so crates.io and GitHub would render 176 dead spans.
• A group label on each FuncId variant. The row knows its group; whether the variant doc should repeat it is a taste call about a page that is already 176 lines long.

Verification

Run here:

• generator cargo test --no-fail-fast — green; backend_suite 311, one of them new.
• cargo clippy --all-targets -- -D warnings — clean on the generator and on the generated crate.
• cargo doc --no-deps -p ta-lib under RUSTDOCFLAGS=-D warnings — clean.
• the crate's 534 doctests, and its --tests suites — green.
• generate leaves the tree clean apart from this diff: no C, Java or C# file moves.

NOT run here: ta_regtest and the cross-language servers. Nothing this touches is compiled into one — the diff is a doc comment and the code that writes it.