…IndexedVector`

`UInt64(std::floor(rhs))` is undefined behavior when `rhs` is negative:
for integer types, `std::floor` implicitly converts to double first;
for float types, casting a negative double to `UInt64` is UB per C++ standard.

`checkValidValue` throws for negative values at runtime, but the compiler
may not see that line 1281 is unreachable for negatives, and UBSan fires.

Fix by using `if constexpr` to skip `std::floor` for integer types and
an explicit non-negative check for float types.

https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=99869&sha=9b37a8c67f4b0fceacf60174b3bb77d2b8477431&name_0=PR&name_1=AST%20fuzzer%20%28amd_ubsan%29

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ric_indexed_vector

…oat scalars

The previous fix addressed the float-to-UInt64 cast for negative values by
going through Int64 first, but the `decimal_value` computation was still
broken for negative floats (converting the two's complement UInt64 back to
float gives a huge positive number, leading to another UB cast).

Rewrite `pointwiseEqual` to use the same unified fixed-point conversion as
`initializeFromVectorAndValue`: compute `Int64 scaled_value` with an explicit
overflow guard, then compare all bits of the two's complement representation
in a single loop. For out-of-range scalars, return an empty bitmap (no element
can match) instead of triggering UB.

Also add test coverage for `pointwiseEqual` with out-of-range and negative
float scalars.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

`static_cast<Float64>(std::numeric_limits<Int64>::max())` rounds up to
2^63 in Float64, so values at exactly 2^63 passed the old `fabs(value) > lim`
guard and then hit `static_cast<Int64>(...)` which is undefined behavior.

Replace the rounded-Float64 bound with an exact-integer-domain check:
compute the scaled value first, then compare against 2^63 (which is
exactly representable in Float64) using `>= int64_upper || < -int64_upper`.

Applied to all three call sites: `initializeFromVectorAndValue`,
`pointwiseEqual`, and `addValue`.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>