[seraphis] ringct: adjust multiexp

[UkoeHB]

This is a PR in my 'upstreaming seraphis_lib project', the changes here are not used anywhere yet.

• Refactors pippenger_cached_data to use a vector of ge_cached directly. This makes pippenger_cached_data reusable (I use it in a multiexp utility in the seraphis_lib).
• Exposes ge_p3 versions of the multiexp functions for efficiency when desired.
• Replaces tabs with spaces in the multiexp notes, so they are legible in text editors that don't render tabs consistently (mine shows all the whitespace in the multiexp notes as single spaces).

[jeffro256]

Since pippenger_p3 takes const&, you can just pass it as is

[jeffro256]

Looks like both straus and pippenger retain the exact same functionality . Would be nice to have some performance tests comparing the _p3 versions to their counterparts to see the efficiency difference. Besides that, just one minor warning thing.

[UkoeHB]

You can check this perf test: https://github.com/monero-project/monero/blob/master/tests/performance_tests/ge_tobytes.h

It's not significant, but still nice to shave off easy wins like this (iirc sarang was a big fan of avoiding serialization/deserialization ops).

[j-berman]

Pointing out for other reviewers there's a nice unit test helping make sure pippenger_cached_data retains behavior here:

monero/tests/unit_tests/multiexp.cpp

Lines 235 to 255 in 50aa0e8

	TEST(multiexp, pippenger_cached)
	{
	static constexpr size_t N = 256;
	std::vector<rct::MultiexpData> P(N);
	for (size_t n = 0; n < N; ++n)
	{
	P[n].scalar = rct::zero();
	ASSERT_TRUE(ge_frombytes_vartime(&P[n].point, rct::scalarmultBase(rct::skGen()).bytes) == 0);
	}
	std::shared_ptr<rct::pippenger_cached_data> cache = rct::pippenger_init_cache(P);
	for (size_t n = 0; n < N/16; ++n)
	{
	std::vector<rct::MultiexpData> data;
	size_t sz = 1 + crypto::rand<size_t>() % (N-1);
	for (size_t s = 0; s < sz; ++s)
	{
	data.push_back({rct::skGen(), P[s].point});
	}
	ASSERT_TRUE(basic(data) == pippenger(data, cache));
	}
	}