speed up getMasses by looking up each distinct element once

[steps-re]

disclosing up front, same as #2262: put together with help from claude, then reproduced, measured and verified by me before opening.

two performance fixes, both in code that runs on every structure load. no behaviour change in either.

1. getMasses looks up every atom individually

getMasses in prody/utilities/misctools.py loops over every atom, calls str.capitalize() twice per atom and does a dict lookup each time. but a structure only contains a handful of distinct element symbols, so nearly all of that work is repeated. switched it to resolve each distinct symbol once and map the result back onto the atoms with unique(..., return_inverse=True).

this is on the path for every pdb, mmcif and mmtf load (pdbfile.py, ciffile.py, mmtffile.py all call it right after parsing elements), so it shows up on essentially any structure a user opens.

measured on my machine, output identical in every case:

structure	before	after
1,079 atoms	0.36 ms	0.03 ms	11.1x
12,750 atoms (antibody)	3.59 ms	0.27 ms	13.3x
100,000 atoms	43.4 ms	5.1 ms	8.5x
1,000,000 atoms	292 ms	45 ms	6.4x

the unknown-symbol case still returns 0.0 rather than raising, the single-string case still returns a scalar, and case-insensitive matching is unchanged.

2. max_len loop in setData

prody/atomic/atomgroup.py finds the longest chain identifier with an explicit python loop. max(map(len, array), default=0) does the same in C, and the default covers the empty case the loop handled by leaving max_len at 0. small, but it runs on every chain assignment.

verification

added TestGetMasses to prody/tests/utilities/test_misctools.py: 7 tests covering known elements, case insensitivity, unknown symbols, repeated symbols (which is what would break if the unique/inverse mapping were wrong), empty input, single-string input and array input. i checked these have teeth by deliberately breaking the mapping and confirming they fail.

ran prody/tests/atomic and prody/tests/utilities: 618 passed. prody/tests/proteins has 16 failures and 1 error on my machine, but they are identical on unpatched main, so they are pre-existing and unrelated to this change (mostly test_insty and a network-dependent mmtf test).

happy to split these into two prs if you'd rather take them separately.

mike