Optimize Format for common verbs without flags

[cuiweixie]

Summary

• Fast-path (*Int).Format for common verbs (%d, %v, %s, %x) when no width, precision, or format flags are set, using Dec() / Hex() instead of allocating via ToBig().
• Fall back to big.Int.Format for all other cases (flags, width, precision, other verbs) to preserve full fmt compatibility, including nil "<nil>" behavior.
• Add BenchmarkFormat and expand TestFormat to compare against big.Int across verbs and flags.

Benchmark (Apple M4, benchstat n=6)

name                 old time/op    new time/op    delta
Format/%d/small-10     162ns ± 3%      66ns ± 8%  -59.36%
Format/%d/large-10     267ns ± 1%     212ns ± 4%  -20.42%
Format/%x/small-10     123ns ± 4%      59ns ± 1%  -52.21%
Format/%x/large-10     138ns ± 5%      78ns ± 0%  -43.63%
Format/%v/large-10     273ns ± 4%     214ns ± 0%  -21.60%
Format/%#x/large-10    148ns ± 3%     158ns ± 1%   +6.51%  (fallback)
Format/%10d/large-10   279ns ± 1%     281ns ± 1%   +0.75%  (fallback)
geomean                188ns          129ns       -31.16%

name                 old allocs/op  new allocs/op  delta
Format/%d/small-10     6.00 ± 0%      2.00 ± 0%  -66.67%
Format/%d/large-10     7.00 ± 0%      2.00 ± 0%  -71.43%
Format/%x/small-10     6.00 ± 0%      2.00 ± 0%  -66.67%
Format/%x/large-10     6.00 ± 0%      2.00 ± 0%  -66.67%
Format/%v/large-10     7.00 ± 0%      2.00 ± 0%  -71.43%
geomean                6.55           2.86       -56.34%

Test plan

• go test -run=TestFormat
• go test -bench=BenchmarkFormat -benchmem -count=6 (before/after via benchstat)
• CI passes on the PR