Fixes extra 40 samples delay with measured directivities (issue #398).

[fakufaku]

With analytical directivity, the fractional delay of the echoes are computed as analytical filter, which necessitates manually adding a global delay so that the fractional delay impulse response fits in the RIR.
With measured directivities, the fractional delays are applied by convolution and naturally contain the global delay.
The global delay was applied to the echo time of arrival in both cases, which resulted in the delay being applied twice for measured directivities.

This fix avoids applying the manual delay when using measured directivities.

• Are there docstrings ? Do they follow the numpydoc style ?
• Have you run the tests by doing nosetests or py.test at the root of the repo ?
• Have you checked that the doc builds properly and that any new file has been added to the repo ? How to do that is covered in the documentation.
• Is there a unit test for the proposed code modification ? If the PR addresses an issue, the test should make sure the issue is fixed.
• Last but not least, did you document the proposed change in the CHANGELOG file ? It should go under "Unreleased".

Happy PR 😃