Fix RampInterpolator double-counting and calc_knownRp pressure conversion

[bernalde]

Summary

• Fix RampInterpolator with count_ramp_against_dt=True and no init key double-counting dt_setpt[0], and clamp negative holdtime to 0
• Fix calc_knownRp.dry early-return path reporting pressure in Torr instead of mTorr
• Fix several test issues: self-comparison assertion, hasattr() on dict, floating-point tolerance, Python <3.11 compatibility

Test plan

• New regression tests for RampInterpolator fix (test_ramp_interpolator_noinit_multisetpt_no_double_count, test_ramp_interpolator_holdtime_clamped_to_zero)
• New regression test for pressure conversion (TestCalcKnownRpEarlyReturn)
• All 140 existing tests pass

Closes #17

🤖 Generated with Claude Code

[Ickaser]

Looked it over. A few suggestions, some comments should be more terse and I don't like introducing tolerances for the ramp interpolation which should not encounter much (if any) numerical noise.

[bernalde]

Addressed the review in 3a75c92. That commit removes the unused fixture, shortens the RampInterpolator comment, makes the early-return mTorr regression check exact, makes the opt_Pch max-bound check exact, and fixes the missing Pchamber["max"] enforcement in opt_Pch_Tsh while tightening the corresponding test. I did not remove the opt_Pch reference-time tolerance in tests/test_opt_Pch.py, because the exact comparison against reference_opt_Pch.csv already differs on this branch by 4.406960061764664e-06 hr (4.238515035519084 vs 4.238510628559022) without a behavioral change in the optimizer. The PR-relevant test slice passes after these changes: 114 passed, 1 skipped from pytest tests/test_functions.py tests/test_calc_knownRp.py tests/test_freezing.py tests/test_main.py tests/test_opt_Pch.py tests/test_opt_Pch_Tsh.py tests/test_opt_Tsh.py.