Refresh the sparse pool, fix CI on SuiteSparse 8, and fix ss.register_new after #630 - #631

Test and allow every sparse release on Python 3.13 and 3.14. The CI pools started sparse at 0.16 on py3.13 and 0.18 on py3.14, and the io extra in pyproject.toml had the same floors. Each was the first sparse release to come out after that Python, on the assumption that older ones might not work there. They do. sparse is noarch, and every release from 0.14 on declares numba >=0.49, numpy >=1.17 and a Python floor of 3.11 or lower (>=3.8 for 0.14/0.15, >=3.10 for 0.16/0.17, >=3.11 for 0.18/0.19), with no ceiling on any of them. Each pin, as conda-forge resolves it (0.14.0, 0.15.5, 0.16.0, 0.17.0, 0.18.0 and 0.19.2), solves on both Pythons and passes the pydata-sparse tests with the oldest and newest numpy/numba pair each Python draws: numpy 2.1 with numba 0.62 and numpy 2.5 with numba 0.67 on py3.13, numpy 2.3 with numba 0.63 and numpy 2.5 with numba 0.67 on py3.14. So every Python now draws from the same pool, 0.14 through 0.19 plus latest, and the io extra requires sparse >=0.14 wherever numba installs. No pairing rules are needed, since the workflow already drops sparse whenever it skips numba. The stress test passes 160000 combinations, and both version audits pass.

Skip the failed-compile JIT test on SuiteSparse 8. Main's CI has been red since #630 merged: every failing job in the latest wave, on main and across the stack, drew psg 8 (SuiteSparse 8.0.2 or 8.2.x, reachable only with numpy 1.x). test_failed_compile_is_not_retried_at_the_next_arming (#630) arms binary.plus on a UDT with C source and makes the compile fail. Ops get C source only where the JIT string setters exist, which is SuiteSparse 9, so on 8.x jit_c_source is None and the test failed at its first assertion wherever the JIT probe passes, which in CI is some macOS runners. It now skips on not _has_jit_set, like the other tests that need an armed op; the stand-in config tests still check on 8.x that _enable_jit_for_udt leaves 'load' alone. Verified on psg 8.0.2.1 and 8.2.1.0 (fails before, skips after) and on 9.3.1.0, the only release in the pool that takes its pre-9.4 branch (runs and passes). The two other SuiteSparse 8 failures are in tests added by #590 and #613, fixed there.

Enable the JIT when registering from C source, and say why it failed. gb.unary.ss.register_new and its binary, indexunary, select, and indexbinary siblings create the op from a C definition and no function pointer, so SuiteSparse has to compile it, or load it from its cache, before the op exists. Under its default jit_c_control of 'run' it does neither, and the call fails with a bare NullPointer. Before #630, touching gb.ss.config first, as these docstrings suggest, raised the control; #630 moved the raise to _enable_jit_for_udt, which these registrars did not call, so each fails in a fresh process on SuiteSparse 8.0.2 through 10.5. CI missed it because the JIT test module turns the JIT on for all its tests. The registrars and gb.dtypes.ss.register_new now create through _compiling_c_source, which raises 'run' (still honoring 'off', 'pause', and 'load') and, when creation fails anyway, raises the same exception again with the reason read off jit_c_control before and after, since SuiteSparse's own error has no message: the JIT is switched off, the control is 'load' and the kernel is not cached, this compile failed, the compiler probe failed in this very call, or no compiler was found. gb.dtypes.ss.register_new also rejects a name of GxB_MAX_NAME_LEN characters up front: the limit counts the NUL, so SuiteSparse cut such a name by one and it failed to compile. New tests cover each registrar in a fresh process, each reason, and the name limit.

Fix small errors in docstrings, messages, and docs. Three __sizeof__ errors printed a literal {backend}; two messages ran words together; the compression-level error omitted "zstd"; the UDT repr size warning ended a sentence twice. Docstrings documented the wrong parameter names (isequal, from_csc, from_dicts, Vector.from_scalar), a nonexistent dtype (Vector.diag), wrong compression defaults and ranges, a graphblas.ops namespace that is graphblas.op, and several parameter lines numpydoc cannot parse. Also a README example that converted the wrong object, a doubled "the", two unclosed quotes, and two CSS comment typos.

Verified locally: full --runslow suites pass on psg 8.0.2.1 and SuiteSparse 10.5.0, both backends, for this branch and for the top of the stack; test_ssjit.py passes on 9.3.1.0 and 9.4.5.0.