build members of keyof mapped types lazily

[Max Schwenk (maschwenk)]

this is stacked on the lazy member tables PR, the first 10 commits here are that one and the last 2 are new. leaving it as a draft until that lands

{ [P in keyof T]: X } mapped types build a symbol for every property of T as soon as anything asks about one of them. in react and test code thats a lot of symbols nobody reads: Partial<Props> just to look up one prop, or vitest's expect(x), whose type extends a mapped type over ~100 chai methods for every distinct x

this gives those mapped types (T is an object type, no as clause) a lazy table:

• a lookup creates just that member, the same way resolveMappedTypeMembers would, and resolving everything later reuses it
• index infos get computed on their own from T's index infos
• mapped types never have signatures, so getSignaturesOfStructuredType doesn't resolve them to find out
• setting up the table does the same steps as resolveMappedTypeMembers up to creating members, so circularity errors show up at the same point (circularBaseTypes.ts caught this)

two places listed every property to answer a narrower question, which would resolve the mapped type anyway:

• somePropertyReducesToNever only cares about names in more than one constituent, so one unresolved constituent gets looked up by name instead of listed. any name it shares with another constituent gets counted from that other one
• lazy member tables no longer list the properties of an intersection base type when they're set up

vs the lazy member tables PR, typescript-benchmarking, median of 3:

	heap, 4 checkers	peak rss, 4 checkers	check, 4 checkers	heap, single	check, single
mui-docs	−18.7%	−24.4%	−21.1%	−11.9%	−12.4%
xstate-main	−2.3%	+1.2%	+1.4%	−2.1%	−5.4%
vscode	−0.1%	+3.0%	+3.1%	−0.1%	−4.0%
webpack	−0.2%	−0.4%	−3.1%	−0.2%	−4.4%
Compiler	+0.4%	−1.2%	−1.4%	0.0%	+3.4%
Compiler-Unions	+0.4%	−1.5%	−9.6%	−0.1%	−0.3%

outside mui-docs the check times are within noise, most of those are under a second

on our 37k-file program heap goes 15.27 → 14.50 GiB with 4 checkers (−5.0%, peak rss 19.1 → 17.7 GiB) and 10.47 → 9.91 GiB single threaded (−5.3%), check time about −5% single threaded. same diagnostics on every run everywhere

full go suite passes (also -race), lint and format are clean. added mappedTypeLazyMembers.ts with baselines generated without this change, so it pins that the output didnt change

• associated issue (same as the lazy member tables PR)
• up to date with main (stacked, see above)
• tests pass (go test ./... in tsc, same thing hereby test runs)
• hereby lint
• hereby check:format
• new tests

used claude code to help write this, ive reviewed it