Fix undefined double-to-int casts when encoding a number value (#5303)

[1820893135-pixel]

Three helpers in ecma-helpers-value.c converted a double to
ecma_integer_value_t and only then tested the result:

ecma_integer_value_t integer_value = (ecma_integer_value_t) ecma_number;

if ((ecma_number_t) integer_value == ecma_number && ...)

The cast is already undefined behaviour when the double is outside the
ecma_integer_value_t range, so the round-trip check cannot guard it. Under
-fsanitize=undefined (the OSS-Fuzz configuration) this traps:

ecma-helpers-value.c:572:40: runtime error: 1.5e+300 is outside the range of
representable values of type 'int'

The same pattern appears in all three functions, and a value such as 1.5e300
walks from one to the next:

function	line
ecma_make_number_value	572
ecma_update_float_number	1021
ecma_value_assign_number	1073

The reported input is a snapshot containing an oversized number literal, which
reaches ecma_make_number_value() through
ecma_snapshot_get_literal(); the value then flows into ecma_update_float_number()
and ecma_value_assign_number(), so fixing only the first site just moves the
trap to the next one.

Fix

Move the range test in front of the cast in each function. Values outside the
range fall through to ecma_create_float_number(), which is what happened for
every value the round-trip check rejected.

Testing

Extends tests/unit-core/test-api-value-type.c with numbers outside the
ecma_integer_value_t range (1e308, -1e308, 1.5e300, -1.5e10, 1e18)
that must round-trip as float values, plus a small integer that must still
behave as before.

Fixes #5303.