Testing

    RestUtils.decodeQueryString("something", "something".indexOf('?') + 1, params);

is not really checking decoding of an empty query.
Instead, it is testing decoding of "something"
as a query (because "something.index('?')+1" evaluates to 0).
The parameter map is left empty because
"malformed" pairs like "something" are currently skipped.

Instead, this change modify this test to check the edge cases:

+ fromIndex >= queryString.length()
+ fromIndex < 0

The code of decodeQueryString() had some trouble with weird URLs:

(1) an input like "uri?param&p=v" causes an exception to be thrown
(2) an input like "uri?param1&param2" causes an infinite loop

This could be verified against an ES server with requests like

    curl -XGET localhost:9200/test/_analyze?t&text=this+is+a+test
    # the exception stack trace shows up in logs

    curl -XGET localhost:9200/test/_analyze?t1&t2&text=this+is+a+test
    # never returns, never ends

This change fixes these issues.

When writing tests for the fix of decodeQueryString() to
handle gracefully edge cases like:

    &a
    &a&b

it arises the question of what behavior is desirable
for these pathological cases (vs the regular p=v pairs).
This change just skips them which is consonant
to the preexisting code.

To be thorough, we add tests for a bunch of edge cases, like:

    QUERY           PARSED PARAMS

    ?               {}
    ?&              {}
    ?=              { "": "" }
    ?a              {}
    ?p=v&a          { "p": "v" }
    ?p=v&a&p1=v1    { "p": "v", "p1": "v1" }
    ?a&b            {}