regression in the representation of exact real balls returned by arb_get_str

[mezzarobba]

Not really a bug, but possibly an unintended side effect of a0f4ad8: With arb 2.22.1, one had

sage: RBF100 = RealBallField(100)
sage: RBF100(10^100)
1.000000000000000000000000000000e+100

whereas with flint 3α, this has become

sage: RBF100 = RealBallField(100)
sage: RBF100(10^100)
[1.000000000000000000000000000000e+100 +/- 1e+65]

[mezzarobba]

In C this is

#include <stdio.h>
#include <flint/arb.h>
#include <flint/fmpz.h>

int main(void) {
    fmpz_t i;
    arb_t a;
    char *str;
    fmpz_init(i);
    fmpz_set_ui(i, 10);
    fmpz_pow_ui(i, i, 100);
    arb_init(a);
    arb_set_fmpz(a, i);
    str = arb_get_str(a, 31, 0);
    printf("%s %d\n",
            str,
            arb_is_exact(a));
}

which outputs

[1.000000000000000000000000000000e+100 +/- 1e+65] 1