Fixed linearized tofts models

[notZaki]

This fix addresses issue #34.
Problem was that the linearized models estimates are [Ktrans, kep, vp], but the package prefers [Ktrans, ve, vp].

I've included the script used to test the models along with the summary statistics at the bottom.
The linear model struggles at very low ve values, which is why it has comparatively poor accuracy on the QIBA 6 phantom (it struggles to fit ve=0.01).
The QIBA 4 phantom has much larger ve values (all ve are >= 0.1), so the linearized models perform about as well as the non-linear approach.

Unrelated change: added the validation's output folder into .gitignore so that I don't have to delete the output before commiting.

---

QIBA 6 - Noiseless
(1321 x 30 points)

	NonLin	Lin
Runtime [s]	1.8	0.4
Kt RMSE	0.059	1.082
Kt errMax	0.123	4.978
Kt CCC	0.9999	0.9996
Ve RMSE	0.082	0.106
Ve errMax	0.234	0.263
Ve CCC	0.9999	0.9999

QIBA 6 - Noisy
(1321 x 3000 points)

	NonLin	Lin
Runtime [s]	15.8	0.99
Kt RMSE	21.33	38.84
Kt errMax	100	100
Kt CCC	0.845	0.659
Ve RMSE	16.14	18.38
Ve errMax	100	100
Ve CCC	0.862	0.835

QIBA 4 - Noiseless
(661 x 90 points)

	NonLin	Lin
Runtime [s]	0.6	0.03
Kt RMSE	6.97	7.05
Kt errMax	23.49	23.7
Kt CCC	0.9998	0.9995
Ve RMSE	18.02	18.03
Ve errMax	99.99	99.99
Ve CCC	0.8904	0.8903
Vp RMSE	23.8	25.23
Vp errMax	92.53	100
Vp CCC	0.9999	0.9999

QIBA 4 - Noisy
(661 x 9000 points)

	NonLin	Lin
Runtime [s]	23.2	1.7
Kt RMSE	11.31	11.4
Kt errMax	100	100
Kt CCC	0.9742	0.9718
Ve RMSE	18.26	18.31
Ve errMax	100	100
Ve CCC	0.7021	0.7008
Vp RMSE	12.68	13.27
Vp errMax	100	100
Vp CCC	0.9717	0.9702

[coveralls]

Coverage remained the same at 0.0% when pulling ce676fc on notZaki:linearModels into 8dd0823 on davidssmith:master.

[davidssmith]

Thanks for this and the .gitignore.