Discount bounded loops in Kotlin; split scaling from repeating

[dejo1307]

The Kotlin extractor emitted a raw loop_depth and no scaling_loop_depth, so perf.scalingDepth() silently substituted the raw value and a constant-count loop inflated the Big-O exponent at full confidence. Kotlin now classifies each loop and emits scaling_loop_depth and calls_in_scaling_loop, joining the Go/Python/TypeScript convention.

Fixing that surfaced a second defect in Go, Python and TypeScript. A single "bounded" predicate drove two different questions: whether a loop adds a factor of n, and whether its body repeats. An infinite loop answers those differently — for {} adds no exponent, but a chain walk inside one issues a query per level. Both are now tracked:

scaling = grows with the input -> scaling_loop_depth
repeats = non-constant trip count -> calls_in_scaling_loop

calls_in_scaling_loop is also emitted whenever calls_in_loop is, even when empty, because the consumer keys off the prop's presence; an omitted key read as "never computed" and fell back to the unfiltered call set. The two defects masked each other — fixing only the emptiness would have dropped true-positive N+1 findings on infinite loops, with green tests and no golden diff, since no fixture covered it.

perf.go needs no logic change; two of its comments described the bug as if it were the design and are corrected.

cacheVersion v97 -> v99, with cachecov entries and golden fixtures pinning all three loop classes in four languages (Ruby and Swift as inline-discount controls).