Fix: explicit N/10 ratings silently dropped by SatisfactionCapture

[rrittich]

Problem

parseExplicitRating() in LifeOS/install/hooks/SatisfactionCapture.hook.ts returns null for the natural N/10 rating form — 10/10, 9/10, thanks, 8 out of 10. The reject-on-slash guard treats the / right after the leading number as invalid input:

const afterNumber = trimmed.slice(match[1].length);
if (afterNumber.length > 0 && /^[/.\dA-Za-z]/.test(afterNumber)) return null; // "10/10" hits this

So every explicit N/10 rating silently falls through to the sentiment-inference path and is recorded with source: "implicit" — a guess — instead of the score the user actually gave. Bare-number and word ratings (8, ten) work; the fraction form does not.

Fix

Add an N/10 / N out of 10 branch before the generic parser (ahead of the reject-on-slash guard), reusing a hoisted SENTENCE_STARTERS guard so a work phrase like 2/10 items done is still correctly not treated as a rating.

Behavior

input	before	after
10/10, thank you	null (→ implicit guess)	{rating: 10, comment: "thank you"}
9/10	null	{rating: 9}
8 out of 10	null	{rating: 8}
2/10 items done	null	null (guarded, not a rating)
8, ten thanks	unchanged	unchanged

No change to existing bare-number or word-form ratings. Single-file, no new dependencies.

[rrittich]

Discovered that when I wanted to give my DA a 10/10 rating, it didn't seem to elevate that to a preference in later chats. So I asked it to investigate and it discovered this issue in the PAI harness' logic. I'm not a programmer, and I have only a faint idea of what it is talking about here, but hopefully it helps and makes sense.