Always inline some tensor functions.

Always inline some tensor functions.

This drastically reduces the time in assembly in ASPECT:

DEBUG:
| Assemble Stokes system          |         8 |      38.3s |        27% |
| Assemble temperature system     |         8 |      51.4s |        36% |
| Build Stokes preconditioner     |         3 |      13.8s |       9.7% |
| Build temperature preconditioner|         8 |     0.192s |      0.13% |
| Solve Stokes system             |         8 |      12.3s |       8.6% |

DEBUG WITH PATCH:
| Assemble Stokes system          |         8 |      30.8s |        25% |
| Assemble temperature system     |         8 |      41.7s |        34% |
| Build Stokes preconditioner     |         3 |      12.1s |       9.9% |
| Build temperature preconditioner|         8 |     0.192s |      0.16% |
| Solve Stokes system             |         8 |      12.2s |        10% |

(Run with 16 processors, using the testcase discussed in https://github.com/geodynamics/aspect/pull/654 .)

There are no significant differences in release mode with this patch,
as probably expected.