PERF: regression in time series plotting

[jorisvandenbossche]

Opening a dedicated issue to not have this lost in #18532.

Using the example from the benchmarks:

N = 2000 
M = 5 
idx = date_range('1/1/1975', periods=N) 
df = DataFrame(np.random.randn(N, M), index=idx)  

I get on master:

In [25]: %%timeit plt.close('all') 
    ...: df.plot() 
                          
2.07 s ± 251 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

but on 0.23.4:

In [2]: %%timeit plt.close('all') 
   ...: df.plot()
   ...: 
87.4 ms ± 979 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

The slowdown was originally even bigger and was addressed already partly #23589, but there is still a ~ 50x slowdown remaining.

[TomAugspurger]

Taking a look at this now.