perf: import awkward/vector/graphviz/particle lazily

[henryiii]

🤖 AI text below 🤖

Summary

import pylhe previously eagerly imported four heavy dependencies that the core reader/writer never needs:

• awkward and vector — only needed for to_awkward (pulled in by the module-bottom from .awkward import to_awkward, which also runs vector.register_awkward() and several ak.mixin_class registrations at import time).
• graphviz and particle — only needed for the LHEEvent.graph visualization (plus the module-level _PDGID2LaTeXNameMap = DirectionalMaps(...) construction).

This PR makes all four lazy so the core parse path imports nearly instantly, with no public API changes.

Changes

• Removed the bottom-of-module from .awkward import to_awkward; to_awkward is now exposed via a module-level __getattr__ that imports it from .awkward on first access (raising the standard AttributeError for unknown names). "to_awkward" stays in __all__/__dir__. from pylhe import to_awkward, pylhe.to_awkward, and import pylhe.awkward all still work, and the vector/awkward behavior registrations now happen only when the awkward bridge is actually used.
• Moved the graphviz/particle imports and the DirectionalMaps construction into LHEEvent._build_graph. The PDGID→LaTeX map is built once and cached via functools.cache. Type annotations are quoted with a TYPE_CHECKING-guarded import graphviz.
• mypy stays green under strict (the # type: ignore[import-untyped] for untyped graphviz now lives on the TYPE_CHECKING import; no stale ignores).

Laziness verification

python -c "import sys, pylhe; assert not any(m in sys.modules for m in (\"awkward\",\"vector\",\"graphviz\",\"particle\"))"

passes after the change. A new tests/test_api.py::test_lazy_heavy_imports asserts this in a fresh interpreter (via subprocess.run) so other tests' imports do not pollute sys.modules. pylhe.to_awkward and LHEEvent.graph were verified end-to-end (covered by tests/test_awkward.py and tests/test_visualize.py).

Notes

• Dependencies are intentionally not changed: graphviz/particle/awkward/vector remain required deps. Making them optional extras is a possible follow-up, deliberately out of scope here.

Part of #378

🤖 Generated with Claude Code

[github-actions]

🏎️ Benchmark Comparison

--------------------------------------------------------------------------------------------------- benchmark: 4 tests --------------------------------------------------------------------------------------------------
Name (time in ms)                                   Min                   Max                  Mean             StdDev                Median                IQR            Outliers     OPS            Rounds  Iterations
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_count_events_benchmark (0002_mr)          420.2988 (1.0)        429.8500 (1.0)        424.6970 (1.0)       2.8136 (1.32)       425.1568 (1.0)       4.1999 (1.65)          2;0  2.3546 (1.0)          10           1
test_count_events_benchmark (0001_base)        423.8510 (1.01)       431.1994 (1.00)       427.5154 (1.01)      2.1313 (1.0)        428.0011 (1.01)      2.5497 (1.0)           3;0  2.3391 (0.99)         10           1
test_fromfile_and_to_awkward (0001_base)     3,493.5209 (8.31)     3,543.3831 (8.24)     3,519.7872 (8.29)     20.3815 (9.56)     3,519.4007 (8.28)     40.6146 (15.93)         5;0  0.2841 (0.12)         10           1
test_fromfile_and_to_awkward (0002_mr)       3,494.8875 (8.32)     3,524.7333 (8.20)     3,510.5020 (8.27)     10.5849 (4.97)     3,508.8153 (8.25)     20.8205 (8.17)          5;0  0.2849 (0.12)         10           1
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Legend:
  Outliers: 1 Standard Deviation from Mean; 1.5 IQR (InterQuartile Range) from 1st Quartile and 3rd Quartile.
  OPS: Operations Per Second, computed as 1 / Mean