Add fixed-point percentile type and integer percentile ranking

[NikolayChirkov]

Summary

Add a reusable fixed-point percentile type for runtime policies and percentile selection without runtime floating-point math.

• Store the complementary tail probability (100 - percentile) in FixedPoint<uint16_t, 100.0>.
• Support compile-time construction for percentiles from p30 through p100.
• Add integer-only percentile-index helpers for fixed-point and integer percentiles.
• Add the one-byte TimeoutFactor8 fixed-point multiplier.
• Reject out-of-range compile-time percentile constants.

Representation

• sizeof(ae::Percentile) == 2
• tail resolution: 1 / 512 percentage point
• sizeof(ae::TimeoutFactor8) == 1
• no runtime float, double, log, pow, or ceil on the percentile/rank path

Representative effective values:

Requested	Stored tail	Effective percentile
p30	70	30
p95	5	95
p99	1	99
p99.9	0.099609375	99.900390625
p99.99	0.009765625	99.990234375
p100	0	100

Validation

• Numeric tests: 238/238 PASS
• Large-N rank tests distinguish p99, p99.9, and p99.99.
• Compile-fail tests cover values below p30 and above p100.

This PR is required by the corresponding aether-client-cpp Presence and cloud-request-policy PR.

[BartolomeyKant]

I didn't get the idea of separated type especially for percentiles.