docs(tools): stop over-quantifying what String does to a big integral double

The pyScalar paragraph read as a universal over every beyond-safe-range
integral number, and three of its clauses have counterexamples inside
that very domain: String(2 ** 53) and String(1e20) are byte-identical to
BigInt's digits, so the "different integer or no integer literal at all"
split is not exhaustive, "the 16 digits" is 2 ** 60's instance count
rather than the mechanism (shortest round-trip is 1 to 17 significant
digits), and padded digits do name a held integer for 1e20. Say shortest
decimal string then padded to the exponent, give both counts, condition
the no-double-holds-it clause, and state the invariant that makes the
rule unconditional: where String is already exact the two agree, and
where it is not, BigInt is the exact one.

Also align one README.zh.md term: the same file already translates
"exotic names" as 特殊名称 in the SDK-section bullet.

packages/core/tools/README.i18n.yaml

@@ -3,4 +3,4 @@
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write packages/core/tools/README.md
 README.md: 20df93e734afb9e7f4280d3aa208af2c8338001c
-README.zh.md: d16a8a90c626c746b8629d148e432302f72b5f30
+README.zh.md: a9741673b7283a78223fb9523abef022a79638e4