Fix NPE for /_cat/indices when no primary shard

[dnhatn]

When a node which contains the primary shard is unavailable, the primary
stats (and the total stats) of an IndexStats will be empty for a short
moment (while the primary shard is being relocated). However, we assume
that these stats are always non-empty when handling _cat/indices in
RestIndicesAction. This commit checks the content of these stats before
accessing.

Closes #26942

[dnhatn]

@jasontedor, I do not like having many null-checks. If there is a better approach, please let me know. Thank you.

[jasontedor]

LGTM. Please add version labels for v5.6.4, v6.0.0, v6.1.0 and backport there too.

[dnhatn]

Thanks @jasontedor. I will backport to v5.6.4, v6.0.0, and v6.1.0.

[jasontedor]

@dnhatn It's fine. Maybe we could do:

    private <T, U> U ifNotNull(final CommonStats stats, Function<CommonStats, T> f, Function<T, U> g) {
        final T x = f.apply(stats);
        return x == null ? null : g.apply(x);
    }

(obviously with better names) and then:

            table.addCell(ifNotNull(totalStats, CommonStats::getSearch, s -> s.getTotal().getSuggestCount()));
            table.addCell(ifNotNull(primaryStats, CommonStats::getSearch, s -> s.getTotal().getSuggestCount()));

at the call sites. It's not clear to me whether or not this is an improvement. Personally I think it's fine as-is, we can keep it simple here. It might be easier to read and maintain. What do you think?

[dnhatn]

@jasontedor, I was trying a slightly different but similar approach to your proposal. However, I found that it was a bit harder to read than null-checks.

[jasontedor]

However, I found that it was a bit harder to read than null-checks.

Yeah, that's what I suspected would be the case.