Add base_indention for cached top-level expression

[lorenzwalthert]

Closes #763.

Here is how the current PR would change benchmark results when merged into master:

cache_applying: 0.03 -> 0.03 (1.7%)
cache_recording: 1.1 -> 1.09 (-0.3%)
without_cache: 3.28 -> 3.26 (-0.5%)

cache_applying: 0.02 -> 0.02 (-2.1%)
cache_recording: 0.74 -> 0.74 (-1.3%)
without_cache: 2.15 -> 2.2 (2.2%)

cache_applying: 0.03 -> 0.03 (-0.5%)
cache_recording: 1 -> 1 (-0.5%)
without_cache: 2.9 -> 2.94 (1.5%)