fix(format): compact counts of a million or more as "1.2M"

[eeshsaxena]

What

formatCount only had a thousands tier, so any value at or above a million rendered as a four-digit thousands string:

formatCount(999_999)     -> "1000k"     (should roll over)
formatCount(1_000_000)   -> "1000k"
formatCount(1_250_000)   -> "1200k"     (should be "1.3M")
formatCount(12_300_000)  -> "12300k"    (should be "12M")

That is neither compact nor readable, and a five-character string like 12300k overflows the fixed-width slots it feeds (the card, the Scout Report metric bars, and the duel comparison rows).

This is reachable in practice: totals for the biggest accounts land in this range, and they get bigger once organization-owned repositories are counted (see #31 / #32) - "stars earned" for a top author is a six-figure number today and clears a million for the very top of the list.

Change

• Add a millions tier that reuses the existing precision rule (one decimal below ten units, none above), so 1_250_000 -> "1.3M" and 12_300_000 -> "12M".
• Roll 999_950+ over to "1M" rather than rounding up to "1000k", so a four-digit thousands value can never be printed.
• Behaviour below a million is unchanged (1240 -> "1.2k", 248723 -> "249k", 9999 -> "10k", 999 -> "999").

Tests

lib/format.ts had no test file, and CONTRIBUTING asks for tests on pure logic, so this adds tests/format.test.ts covering formatCount (small values, the thousands tier, the decimal-drop at ten thousands, the new millions tier, and the rollover boundary) plus round1 / round2.

• The two new millions assertions fail on master (expected '1000k' to be '1M') and pass with this change.
• Full suite green: 18 files, 168 tests. npm run lint reports 0 errors.

An AI assistant helped spot and implement this; I reviewed the change and ran the tests and linter locally.

[Younesfdj]

Hello @eeshsaxena thanks for your interest in contributing to gitfut, good catch