lite2: non-recursive bisection may be slower than recursive version #4504
Description
while this is true for the example you show in Fig.1, not sure this is always the case. For example, in another case you can have something like this:
start with th=1, ih=256
the +1/3rd check fails a few times and interim headers are retrieved and checked, let's say: [1], 8, 16, 32, 64, 128, [256]
8 verifies against 1 so now we have th=8, ih=256
again the +1/3rd check fails and other headers are retrieved: [8], 16, 24, 40, 72, 136, [256]
16 verifies against 8 and at this point th=16, ih=256
-...
The downloaded headers are not kept around (so we save on memory) but more headers and vals need to be retrieved.
See #4496 (comment) for the whole discussion.