Hi, sRGB imac will be display characteristics, basically a slightly different gamma curve at the base affecting the shadows area compared to rec709. sRGB and rec709 share same color primaries that define 'width' of gamut in the wider scheme of things.
Working space is to do with exactly that, the space in terms of gamut defined by the choice of color space, to manipulate your color values within, wide enough to generally avoid clipping gamut unnecessarily leading to such things as potential anomalies in final image 'quality.'
A wider gamut working space is more of a possibility when working with source files of greater bit depth than 8bit. But decision whether to bother also depends on final delivery ie: rec709 video.
The principle of color management in the case of apps like AE is that you define or in AE terms 'interpret as' your source files color space ie: rec709 for video, sRGB for images or camera raw spaces, set a working space gamut as wide as or a little wider than input sources or in the case of raw where color space is undefined and at greater than 8bit a choice of work space based on widest gamut output you intend going to, in your case rec709 in DNxHD.
Then depending on the width of the gamut of your work space you preview either as rec709 if viewing through a HDTV for example or sRGB monitors depending on your display device disregarding projectors and AE will do the necessary transform and dither from wider working space gamut and bit depth to display space so you see what you'll get in your 8 or 10bit video encode or close to it depending on a host of other factors concerning playback, codecs and the calibration minefield.
Regarding bit depth rec709 doesn't mean 8bit, the two are not related, yes rec709 defined video can be 8bit or 10 or 16bit in YCbCr.
If you encode raw 14bit into 16bit then you have 16bit levels range, whether that's necessary or depending on the source files bit depth worthwhile considering storage cost is another thing. 14bit raw precision into linear RGB minus noise floor setting black level around 4000 to 5000 and sensor saturation of first channel with the other channels scaled accordingly ie: RGB multipliers to get your WB right, will undoubtedly mean not even 14bits worth of levels. Black level (5000) will be set to 0 in 16bit and sensor saturated channel say approx 13584 for a 550D, scaled to 65535 in 16bit.
8bit from raw is not enough really for intermediate storage ie: DNxHD Prores etc, 10bit log probably would be, 16bit linear or gamma encoded more than enough.
