chore: remove every comment from the Rust sources

[LeadcodeDev]

Stacked on #343 — targets feat/install-cli-and-studio, not main. Retarget to main once #343 merges.

16 355 comments removed across 199 changed files. // and //! are both at
zero. The 1 671 /// that remain are the ones clap and schemars read.

	before	after
//	6 298	0
//!	1 255	0
///	8 943	1 671 (clap + schemars only)

Why those 1 671 stay

They are not documentation, they are interface text:

Read by	Where	What it produces
clap	types deriving Parser/Subcommand/Args/ValueEnum	the text of rustmotion --help
schemars	types deriving JsonSchema, and their fields	the description fields of the exported schema, which validate --strict-attrs consumes

Removing them would not delete a comment. It would empty --help and strip
1 223 descriptions out of the schema.

Done by a lexer, not a regex

"https://example.com/a//b" contains //. A raw string can hold
/* anything */. The stripper tracks string, raw-string, char and
block-comment state, so no literal was touched.

Two invariants, and they both caught real losses

Captured from a binary built before the change and diffed against one built
after:

• rustmotion schema → byte-identical, 1 223 descriptions preserved
• rustmotion --help and all twelve subcommands → byte-identical

Both failed on the first attempt, which is why they were captured. Two
structural faults in the detector:

A multi-line attribute. #[derive(Parser)] followed by a #[command(…)]
spanning five lines: the inner lines (name = "rustmotion",) match neither
"attribute" nor "doc comment", so they read as code and cleared the pending
derive list before struct Cli was reached. Every global option lost its help:

-  -q, --quiet              Suppress all output except errors
+  -q, --quiet

Braces inside string literals. Comments were blanked before counting brace
depth; literals were not. A { inside an about = "…" or a format! shifted
the depth that decides where a protected struct body ends, so fields near the
end of a struct fell outside it — six badge fields, counter's duration,
heatmap's animate and others lost their schema descriptions.

Three consequences that are not cosmetic

Five doctests were tests, and they lived in doc comments

VarScope, build_css2_url, parse_ttf_urls, family_slug and expr's
module example. Ported to #[test] functions in the mod tests each file
already had, rather than lost.

Build	Tests before	Tests after
default	1510	1510
--features studio	1701	1701

Two if chains had identical branches — only the comments differed

let depth_sin = if tilt_rad.abs() > 0.01 {
    // With tilt, depth is based on the untilted Y
    theta.sin()
} else {
    // Without tilt, use Y component for depth
    theta.sin()
};

stepper.rs had the same shape: "Filled circle for active" and "Filled circle
for completed" are the same call. Both collapsed to exactly what they already
computed, so nothing renders differently — but the comment was claiming a
distinction the code never made, and collapsing makes that visible instead of
hidden. Worth a second look from whoever knows what the tilt branch was meant
to do; I deliberately did not invent it.

One flaky test, pre-existing

annotation_id() derived an id from SystemTime::now().as_nanos() alone, so two
annotations minted inside one clock tick collided — a real defect in the studio's
id generator, not just a capricious test. It now carries an atomic tiebreaker;
40 consecutive runs green.

Not caused by this PR and not by #343 either: the function is byte-identical to
before the strip, and the studio already built at opt-level = 3 under its old
package. Running the full suite is simply what tripped it.

Verification

• cargo fmt --all --check clean
• cargo clippy --workspace --all-targets -- -D warnings clean, and with --features rustmotion/studio
• both test configurations at their original counts
• the two byte-identical invariants above

CLAUDE.md

The rule was crate-scoped to the studio; it now covers the whole codebase, names
the clap/schemars exception as interface text rather than an exemption, and
records the non-obvious corollary: a doctest lives in a doc comment, so writing
an executable example from here on means writing a test.

[LeadcodeDev]

Rebased onto main now that #342 and #343 are in.

The conflict was structural rather than semantic: this branch carried the install
work as its own commit, while main received the squashed version of it, so the
two histories described the same change differently. Replaying only the strip
commit (git rebase --onto origin/main <install-commit>) let git recognise the
install work as already applied and skip it, which left exactly one real
conflict: gradient_text.rs, the file #342 touched.

Resolved by taking main's version — the #337 fix with its comments — and
running the stripper over it, so the new code stays and only its comments go. 84
comments removed from that file.

Both invariants re-verified against the same pre-strip baseline, which also
confirms neither #342 nor #343 moved them:

• rustmotion schema — byte-identical, 1 223 descriptions
• rustmotion --help and all twelve subcommands — byte-identical

Test count rises by exactly the four tests #342 added:

Build	before	now
default	1510	1514
--features studio	1701	1705

All ten gradient_text tests pass, so the text-align fix survives the strip.
fmt needed one pass after the rebase — the lines it wrapped around the removed
comments re-fold once they are gone — and is folded into the commit. clippy
clean in both configurations.