handle 'worlds' for scales more thoughtfully

[clarkedavida]

Per Claude:

Split the one world into the three you identified, and stop deriving the physical-units one from
Nf:

1. Nf (ensemble sea content) — already there; now accepts None = pure gauge. Drives mass/LCP
   handling and the default year for the a(β) parametrization.
2. aWorld — world the a(β) fit was done in. Already implicitly chosen by the year arg to r0_div_a
   / a_div_r1 / a_times_fk (2017 ⇒ Nf0, 2012 ⇒ Nf21…). Make it explicit: have those functions
   carry the Nf they were fit at, store it as self.aWorld, print it in paramSummary, and warn when
   aWorld != Nf instead of silently mixing.
3. physWorld — world the scale was fixed in fm. A property of r0_phys/r1_phys/fk_phys, not of your
   ensemble. r0 and r1 each have exactly one determination (Nf21), so in setScales just call
   r0_phys() / r1_phys() with no world= and delete the try/except world=self.world dance for them.
   Keep a real world selector only for fk, where genuinely different Nf determinations exist
   (nature / Nf21 / …).

Result: scaleType='r0' with no Nf works cleanly, the CHECKWORLD fallback machinery disappears for
r0/r1, and paramSummary shows e.g. Nf = None · a-world = Nf0 (r0/a 2017) · phys-world = Nf21 (r0 =
1.5092·r1, MILC 2010) — the hybrid is visible rather than hidden.