ja: read the function-application of as オブ

[yasumorishima]

Follows #730 on the ja branch. Same defect as the absolute value in #725 and the large operators in #726, but in the generic rules, so it covers every function application at once.

Rebased onto ja now that #728, #730 and #733 have landed, and extended to the sites I was holding back while they were open.

の does not mean "of" here

sin x was read サイン の x. Japanese の attaches the other way round, so that parses as "the sine belonging to x" — the operand ends up modifying the function name rather than being applied to it. The same reading appeared for roots (ルート の x) and for any uncaught intent through function-intent.

Yamaguchi, Kawane and Sawazaki (1996) introduce オブ as a borrowed preposition for exactly this position (item 3), and item 6 spells the readings out: ログ底 a オブ x for a logarithm and サイン・インバース（オブ）x for an inverse function. Item 1 says not to reorder into Japanese word order, so the fix is to keep the order and swap the connective.

There are two paths, not one

Changing SharedRules/default.yaml fixed the gradient but not the sine. The sine goes through the invisible function-apply character instead: unicode.yaml gives U+2061 a conditional reading when it follows a trig function or a geometry shape. Both paths are オブ:

• SharedRules/default.yaml — root, root-with-index, apply-function, function-intent
• unicode.yaml — U+2061

Three sites had the connective fused into the word

Grepping for a standalone t: "の" does not find them, which is why they survived the first pass: ∆ and √ in unicode.yaml, and the square-root-of prefix in definitions.yaml. en reads laplacian of and square root of as two words in the same places.

The square root in ClearSpeak_Rules.yaml, SimpleSpeak_Rules.yaml, SharedRules/general.yaml and overview.yaml is the same defect; those files were in #730, so they are here rather than in a conflict.

SharedRules/calculus.yaml is dormant, and fixed anyway

Neither ja nor en includes that file — only fi, es, vi, sv, id and zh-tw do. Its four vector calculus operators reach speech through the function= entries in definitions.yaml instead, and that is what vector_calculus_operators covers. Two of the four, divergence and gradient, had lost the connective in calculus.yaml altogether while laplacian and curl beside them kept theirs, so I brought all four in line rather than leave a file that is wrong the day someone includes it. Please say if you would rather I dropped that hunk.

What was deliberately left alone

の is correct Japanese in most of the places it appears and must stay. The exponent reading added in #721 — 3 の 2 乗, "3 to the 2nd power" — is ordinary Japanese, as are 分の in fractions and 正の / 負の. A blanket replacement would have broken all of them, so each site is matched individually.

順列の in SharedRules/general.yaml is left for now. en reads permutations of there, but the rule already reverses its two arguments, so the right Japanese wording is a decision rather than a swap, and I would rather not fold a judgement call into a mechanical one.

unicode-full.yaml is untouched, since you mentioned in #715 that it can be seeded from the MathPlayer Japanese data.

Checks

audit-translations ja is unchanged against the current ja: 3674 untranslated, 28 rule differences, on both. Test expectations are the actual CI output, including the comma the engine inserts in コタンジェント, オブ x.